tp钱包中国官方网站|mssf

首页
mssf

tp钱包中国官方网站|mssf

作者： tp钱包中国官方网站

2024-03-07 20:02:37

Maruti Suzuki Smart Finance - Car Loans for Alto, Wagon R, Swift, Dzire, Ertiga, Vitara Brezza, S-PRESSO, Celerio, Eeco

Home

Corporate

About Us

History

Leadership

Strengths

Values

Exports

Sustainability

CSR

Investors Disclosure under Reg. 46 of SEBI (LODR) Regulations

Details of the Business

Company Reports

Financials

News

Stock Information

Codes & Policies

Forms and Circulars

Events

Committees

Terms of Appointment of Independent Directors

Contacts

Careers

Life at MSIL

Why Work with Us

Join Us

Meet Our People

Training Academy

Fraudulent Recruitment

Media

Press Releases / Stock Exchange Updates

Together against COVID-19

Maruti Suzuki world

Events

SUV Experiences

Technology

Performance and Fuel Efficiency

S-CNG

Hybrid

Automatic

Suzuki Connect

Safety

Infotainment System

Reach Us

All Offices

Business Opportunities

Locate a Dealer

Business Payments to MarutiSuzuki

Sales

MARUTI SUZUKI ARENA

All Cars

Hatchbacks

Alto K10

Alto

S-PRESSO

Celerio

Wagon R

Swift

Explore

Sedans

Dzire

MUVs / SUVs

Ertiga

Brezza

Vans

Eeco

S-CNG

Alto

Alto K10

Brezza

Celerio

Dzire

Ertiga

Eeco

Swift

S-Presso

WagonR

Explore

Tour

TOUR H1

TOUR H3

TOUR S

TOUR V

TOUR M

Accessories

Maruti Genuine Accessories

Genuine Accessories

Rewards

Maruti Suzuki Rewards

An all-new loyalty program where customers are treated like family

ExploreApply Now

Arena World

Accelerate into the world of Arena with the latest updates, news and information on Maruti Suzuki Arena Cars.

Explore

NEXA

TRUE VALUE

COMMERCIAL

Service

搜狗皮肤编辑器 for Mac

手机版

PC版

编辑器

拼音输入法

五笔输入法

皮肤

更新日志

搜狗皮肤编辑器for Mac 1.0.0

运行环境：Mac OS X 10.10 或更高版本

立即下载

升级日志

搜狗皮肤编辑器for Mac 1.0.0

2019.02.28

新增

1. 皮肤文件升级为全新.mssf格式。

2. 支持可视化的候选项，背景图，特征图，翻页按钮实时调整。

3. 独家背景图阴影设置，让你的作品更Mac。

4. 自定义通知样式，包括大量独家动画样式。

5. 支持Retina级别的皮肤本地检测。

6. 支持原有皮肤后台账户，本地一键发布至皮肤后台。

修复

1. 修复OSX10.14下闪退的问题；

2. 修复皮肤编辑器无法提交的问题；

问下官方皮肤是 .ssf 的这个.mssf要怎么安装双击没有应用程序能打开 · Issue #1 · xiaochunjimmy/Sogou-Input-Skin · GitHub

Toggle navigation

Product

Actions

Automate any workflow

Packages

Host and manage packages

Security

Find and fix vulnerabilities

Codespaces

Instant dev environments

Copilot

Write better code with AI

Code review

Manage code changes

Issues

Plan and track work

Discussions

Collaborate outside of code

Explore

All features

Documentation

GitHub Skills

Blog

Solutions

For

Enterprise

Teams

Startups

Education

By Solution

CI/CD & Automation

DevOps

DevSecOps

Resources

Learning Pathways

White papers, Ebooks, Webinars

Customer Stories

Partners

Open Source

GitHub Sponsors

Fund open source developers

The ReadME Project

GitHub community articles

Repositories

Topics

Trending

Collections

Pricing

Search or jump to...

Search code, repositories, users, issues, pull requests...

Clear

Search syntax tips

Provide feedback

We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Cancel

Submit feedback

Saved searches

Use saved searches to filter your results more quickly

Name

Query

To see all available qualifiers, see our documentation.

Cancel

Create saved search

You signed in with another tab or window. Reload to refresh your session.

You signed out in another tab or window. Reload to refresh your session.

You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

xiaochunjimmy

Sogou-Input-Skin

Public

Notifications

Fork

102

Star

1.3k

Code

Issues

Pull requests

Discussions

Actions

Security

Insights

Additional navigation options

Code

Issues

Pull requests

Discussions

Actions

Security

Insights

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Pick a username

Email Address

Password

By clicking “Sign up for GitHub”, you agree to our terms of service and

privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?

to your account

Jump to bottom

问下官方皮肤是 .ssf 的这个.mssf要怎么安装双击没有应用程序能打开

Closed

eleanors opened this issue

May 29, 2019

· 9 comments

Closed

问下官方皮肤是 .ssf 的这个.mssf要怎么安装双击没有应用程序能打开

eleanors opened this issue

May 29, 2019

· 9 comments

Comments

Copy link

eleanors

commented

May 29, 2019

No description provided.

The text was updated successfully, but these errors were encountered:

All reactions

Copy link

Owner

xiaochunjimmy

commented

May 29, 2019

.mssf好像是官方针对苹果Mac版本输入法出的皮肤格式，理论上双击后就安装并且切换成功了。你是用的Mac吗？

All reactions

Sorry, something went wrong.

Copy link

Author

eleanors

commented

May 29, 2019

.mssf好像是官方针对苹果Mac版本输入法出的皮肤格式，理论上双击后就安装并且切换成功了。你是用的Mac吗？

是Mac 输入法是搜狗五笔

All reactions

Sorry, something went wrong.

Copy link

Owner

xiaochunjimmy

commented

May 29, 2019

这个是搜狗拼音输入法的皮肤

All reactions

Sorry, something went wrong.

Copy link

Owner

xiaochunjimmy

commented

May 29, 2019

因为我没有用五笔输入法，所以实在不确定是不是可以。

https://pinyin.sogou.com/mac/skineditor.php 从这个页面来看，最下面有提到ssf升级为mssf，所以不清楚是不是只针对拼音，你试试从输入法的设置界面皮肤外观那块手动添加，看能成功不

All reactions

Sorry, something went wrong.

Copy link

Author

eleanors

commented

May 29, 2019

这个也试过还是不行识别不了.mssf格式的输入法是最新版本的

All reactions

Sorry, something went wrong.

Copy link

Owner

xiaochunjimmy

commented

May 29, 2019

好吧，可能确实只支持拼音输入法

All reactions

Sorry, something went wrong.

Copy link

J3n5en

commented

May 29, 2019

我这边是搜狗输入法（5.3.0.7499，提示了很多次，没升级.

双击也是打不开，不过去设置——外观里面手动导入是ok的，遇到同样问题的可以试试。

All reactions

Sorry, something went wrong.

Copy link

Przeblysk

commented

May 29, 2019

楼上正解

All reactions

Sorry, something went wrong.

Copy link

Owner

xiaochunjimmy

commented

May 29, 2019

@J3n5en @Przeblysk

感谢楼上提醒，我更新一下文档

All reactions

Sorry, something went wrong.

xiaochunjimmy

mentioned this issue

May 30, 2019

Mac 搜狗五笔不能使用该皮肤

Closed

xiaochunjimmy

closed this as completed

Nov 29, 2020

to join this conversation on GitHub.

Already have an account?

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

No branches or pull requests

4 participants

Footer

Footer navigation

Terms

Privacy

Security

Status

Docs

Contact

Manage cookies

Do not share my personal information

You can’t perform that action at this time.

Sogou-Input-Skin: 搜狗输入法皮肤

开源

企业版

高校版

搜索

帮助中心

使用条款

关于我们

开源

企业版

高校版

私有云

Gitee AI

NEW

我知道了

查看详情

代码拉取完成，页面将自动刷新

捐赠

捐赠前请先登录

取消

前往登录

扫描微信二维码支付

取消

支付完成

支付提示

将跳转至支付宝完成支付

确定

取消

Watch

不关注

关注所有动态

仅关注版本发行动态

关注但不提醒动态

Star

Fork

xiaoxiangyuan / Sogou-Input-Skin

代码

Issues

Pull Requests

Wiki

统计

流水线

服务

Gitee Pages

JavaDoc

PHPDoc

质量分析

Jenkins for Gitee

腾讯云托管

腾讯云 Serverless

悬镜安全

阿里云 SAE

Codeblitz

我知道了，不再自动展开

加入 Gitee

与超过 1200万开发者一起发现、参与优秀开源项目，私有仓库也完全免费：）

免费加入

已有帐号？

立即登录

该仓库未声明开源许可证文件（LICENSE），使用请关注具体项目描述及其代码上游依赖。

项目仓库所选许可证以仓库主分支所使用许可证为准

master

管理

分支 (1)

master

克隆/下载

HTTPS

SSH

SVN

SVN+SSH

下载ZIP

登录提示

该操作需登录 Gitee 帐号，请先登录后再操作。

立即登录

没有帐号，去注册

提示

下载代码请复制以下命令到终端执行

为确保你提交的代码身份被 Gitee 正确识别，请执行以下命令完成配置

git config --global user.name userName

git config --global user.email userEmail

初次使用 SSH 协议进行代码克隆、推送等操作时，需按下述提示完成 SSH 配置

生成 RSA 密钥

获取 RSA 公钥内容，并配置到 SSH公钥中

在 Gitee 上使用 SVN，请访问使用指南

使用 HTTPS 协议时，命令行会出现如下账号密码验证步骤。基于安全考虑，Gitee 建议配置并使用私人令牌替代登录密码进行克隆、推送等操作

Username for 'https://gitee.com': userName

Password for 'https://userName@gitee.com':

私人令牌

新建文件

新建 Diagram 文件

新建子模块

上传文件

分支 1

标签 0

贡献代码

同步代码

创建 Pull Request

了解更多

对比差异

通过 Pull Request 同步

同步更新到分支

通过 Pull Request 同步

将会在向当前分支创建一个 Pull Request，合入后将完成同步

章萧醇

Merge pull request #22 from jrfeng/master

e9781f7

38 次提交

提交

取消

提示:

由于 Git 不支持空文件夾，创建文件夹后会生成空的 .keep 文件

for_windows

保存

取消

Boundary.mssf

保存

取消

Carbon.mssf

保存

取消

Graphite.mssf

保存

取消

Matrix.mssf

保存

取消

README.md

保存

取消

SystemSkins.plist

保存

取消

Tangerine.mssf

保存

取消

Tron.mssf

保存

取消

README

自用的搜狗拼音输入法皮肤，重新对素材进行了无损压缩，调整了细节，分享给大家。

2021年03月07日更新新增 Carbon 碳黑皮肤风格；

2020年11月29日更新在 v1.3 中修复了苹果系统更新导致的 5 款皮肤候选项和翻页箭头不居中的问题，感谢 Hugh Sun的反馈并邮件提供了部分调整支持！❤️

2019年6月14日更新新增 Windows 版输入法皮肤，预览图，感谢 jrfeng ！❤️

设计理念

由于原本是为了自用，所以在风格和理念上都是按照我自己的喜好来的：简洁纯净，剥离过分设计带来的视觉纷扰，享受打字时的简单和愉悦...

1. Tron 创

Tron 是我使用最久的一款皮肤，名称源自于我非常喜欢的一部视觉大片《Tron: Legacy》(中文名称《创: 战纪》)，如果用两个颜色来代表科技感，那么它一定是蓝色和白色，正如《Tron》系列。色彩上采用饱和度较高的蓝色，彰显其个性与特点。

2. Tangerine 橘色

Tangerine 名称取自于同名电影《Tangerine》(中文名称《橘色》)，比较值得一提的是，本电影全程用 iPhone5s 拍摄，后期通过手机 App 进行调色。并不是因为导演像陈可辛、贾樟柯一样在给苹果打广告，是因为导演真穷，买不起专业设备，可能连防抖云台都买不起，所以画面也有点抖。题材比较敏感，剧情不好评价，感兴趣的小伙伴自己感受一下。

3. Graphite 石墨

Graphite 石墨是最近才制作的皮肤，也是我最近一直在用的皮肤，它的设计初衷就是“尽可能简单”，希望最终呈现给用户的感受是“没有设计”，就像毫尖蘸墨轻拂宣纸一般自然纯粹，配色采用黑与白（考虑对比度过于强烈带来的视觉疲劳，没有采用纯黑），黑如墨，白如纸，取名石墨。

4. Boundary 界线

Boundary 是一款无阴影的皮肤，文字候选框与背景在视觉上紧密贴合，采用深色边框构成界线以区分候选框与背景，即使在色彩杂乱的背景上亦能清晰定位内容焦点，构成视觉饱和的同时不失简洁优雅，甚至还稍微有点复古。设计思路来自 Testdog 同学的建议。

5. Matrix 矩阵

Matrix 名称和风格取自同名电影《Matrix 黑客帝国》，因此风格设定也比较有极客色彩。严格来说，这并不是一款针对夜间模式的皮肤，一方面搜狗输入法并没有针对页面模式的皮肤自动适配能力，而为了夜间模式手动切换输入法皮肤其实也是一个比较反人类的交互行为；另一方面很多用户即使在白天也喜欢使用 Dark Mode 深色模式，因此皮肤的配色并不是完全针对夜间场景进行适配的，稍微加强了对比度和饱和度，又重新调整了黑客帝国代码绿的色相，使得皮肤看起来更具设计感。

6. Carbon 碳黑

Carbon 碳黑，参照石墨风格制作的深色模式，喜欢石墨又喜欢用深色模式系统的同学可以试试，还不是很成熟，有时间再慢慢优化；

使用方法

方法一：下载本项目，Mac 版用户请找到 mssf 文件，双击即可完成皮肤安装和切换。

Windows 版请在 for_windows 文件夹下找对应的 ssf 文件。

方法二：如果方法一没有成功，可以通过打开搜狗输入法的的 [偏好设置]，然后在最顶部 Tab 栏选择 [外观]，然后点击左下角的 [+] 加号按钮，从本地目录里选择 mssf 文件就可以了，也可以直接拖动 mssf 皮肤文件直接到 [外观] 设置项的面板中（感谢 J3n5en 同学在 issue 中提醒）。

附：下拉候选项样式异常解决办法

安装过后，你可能会发现使用 Mac 搜狗输入法的卷轴模式的话，会出现样式异常问题，类似下图（以Boundary 皮肤为例）：

搜狗输入法设定为：只有在皮肤商店上架（白名单）的皮肤在 “卷轴模式” 下才能完美展示，否则下拉卷轴展示效果稍微不太美观；而上架皮肤商店需要审核，到目前为止还没有收到审核通过的消息。所以如果你希望本皮肤在 “卷轴模式” 下完美展示，可以通过手动修改本地白名单的方式来进行。

Mac 修改白名单方法：

打开 finder，然后在 finder 浏览器下按 Command + Shift + G 打开跳转窗口，然后在文本框内输入 /Library/Input Methods/SogouInput.app/Contents/Resources 后敲回车，就打开了搜狗输入法的资源文件目录。

将资源目录中的 SystemSkins.plist，替换为本项目列表中的同名文件 SystemSkins.plist，替换后重新切换一下皮肤即可。

本文件是在最新的（2020年11月29日）白名单皮肤基础上，添加了以上几个皮肤。其他没有修改。或者你也可以提前备份一下原来的文件，自己手动添加以上皮肤。

替换上述文件并重新切换激活皮肤之后，卷轴模式下拉候选项的样式应该变成这样了：

设计相关

剩下这部分内容给热爱折腾的朋友，如果你也想做属于自己的皮肤：

设计工具：Sketch

用于绘制和导出皮肤需要的基本素材，你也可以使用Photoshop或Illustrator等其他设计软件。

压缩工具：image-optim

用于压缩设计工具导出的图片素材，Sketch 在无插件情况下，默认导出的 PNG 是没有经过高级压缩的，而PNG格式的图片可以在压缩工具下再次进行无损压缩，降低皮肤体积，经测试，压缩率在50%左右。

制作工具：搜狗皮肤编辑器 for Mac

搜狗官方皮肤编辑工具，目前还有点小bug(v1.0.0)，似乎几年未更新了[‍♀️]

可以使用上面的皮肤编辑器打开我的皮肤.mssf文件，在此基础上二次调整。也可以参考我的皮肤尺寸进行素材的设计。

如果还有什么疑问，或者有什么建议，可以在 GitHub 上创建 issue来反馈提问。

精神支持

如果你喜欢我的皮肤，可以在 GitHub 本项目右上角点一下 Star 来支持我（Star 一般用于支持，也常被 GitHub 用户作为收藏操作），开源精神也需要精神支持。

空文件

Starred

Star

Fork

捐赠

0 人次

举报成功

我们将于2个工作日内通过站内信反馈结果给你！

请认真填写举报原因，尽可能描述详细。

举报类型

请选择举报类型

举报原因

取消

发送

误判申诉

此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。

如您确认内容无涉及不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。

取消

提交

简介

搜狗输入法皮肤

展开

收起

暂无标签

保存更改

取消

发行版

暂无发行版

贡献者

全部

近期动态

加载更多

不能加载更多了

编辑仓库简介

简介内容

搜狗输入法皮肤

主页

取消

保存更改

https://gitee.com/xiaoxiangyuan/Sogou-Input-Skin.git

git@gitee.com:xiaoxiangyuan/Sogou-Input-Skin.git

xiaoxiangyuan

Sogou-Input-Skin

master

Git 大全

Git 命令学习

CopyCat 代码克隆检测

APP与插件下载

Gitee Reward

Gitee 封面人物

GVP 项目

Gitee 博客

Gitee 公益计划

Gitee 持续集成

OpenAPI

帮助文档

在线自助服务

更新日志

关于我们

加入我们

使用条款

意见建议

合作伙伴

售前咨询客服

技术交流QQ群

微信服务号

client#oschina.cn

企业版在线使用：400-606-0201

专业版私有部署：

13670252304

13352947997

开放原子开源基金会

合作代码托管平台

违法和不良信息举报中心

粤ICP备12009483号

简体

繁體

English

点此查找更多帮助

搜索帮助

Git 命令在线学习

如何在 Gitee 导入 GitHub 仓库

Git 仓库基础操作

企业版和社区版功能对比

SSH 公钥设置

如何处理代码冲突

仓库体积过大，如何减小？

如何找回被删除的仓库数据

Gitee 产品配额说明

GitHub仓库快速导入Gitee及同步更新

什么是 Release（发行版）

将 PHP 项目自动发布到 packagist.org

仓库举报

回到顶部

登录提示

该操作需登录 Gitee 帐号，请先登录后再操作。

立即登录

没有帐号，去注册

P3 | msater-slave sampling filter (MSSF) (JSSC-2016-03)_a 2.4-ghz 6.4-mw fractional-n inductorless rf synt-CSDN博客

P3 | msater-slave sampling filter (MSSF) (JSSC-2016-03)

最新推荐文章于 2023-07-21 08:07:34 发布

Clara_D

最新推荐文章于 2023-07-21 08:07:34 发布

阅读量700

点赞数

分类专栏：

论文

文章标签：

本文链接：https://blog.csdn.net/Clara_D/article/details/117399728

版权

论文

专栏收录该内容

14 篇文章

8 订阅

订阅专栏

今天看到一篇2016的JSSC，讲的是用Ring VCO的TypeI型PLL中滤波器结构的改进，从而带来了性能的提升。

MSSF

MSSF原理

如下图所示，Fig.3是传统的滤波器结构，这种结构的Vctr的时域波形会有波动，给PLL带来spur。 Fig.4是该文提出的新结构MSSF：msater-slave sampling filter，利用这种sample and hold的结构可以保持Vctr在环路锁定时输出值恒定。从而降低了杂散。该滤波器的传输函数如下：其中，Req表示两个开关与C1的等效电阻。如果把这个滤波器当成零阶保持器(zero-order hold (ZOH) circuit)来看待，可以得到更准确的传输函数如下：

MSSF对相位裕度和环路带宽带来的影响

使用MSSF的锁相环对相位裕度带来的影响（一堆等效化简……）：此时的等效环路带宽可以扩大，解释如下，闭环传输函数为（其中H(jw)使用上图中的简化版）：从而可以让环路带宽在

0.55

−

0.71

0.55f_{REF}-0.71f_{REF}

0.55fREF−0.71fREF范围内也能保持较好的稳定性，留足相位裕度。

谐波抑制

除此之外，为了进一步减小杂散（比如来自变容管leakage current、开关S2 leakage current, charge injection产生的Vctr ripple），文章还采用了谐波抑制来减少杂散。利用回转器(gyrator)/有源电感(active inductor)，可以将谐波抑制电路等效为如下图所示的电路，如果假设输出阻抗为无穷，此时

(

)

Z_{in}=L_{eq}s=C_{L}s/(G_{m1}G_{m2})

Zin=Leqs=CLs/(Gm1Gm2)，

G_{m1}

Gm1和

G_{m2}

Gm2组成的电路就是回转器。这种有源电感具体分析思路可见The Active Inductor 最后得到的PLL设计及性能如下图所示，谐波抑制电路在提供抑制杂散的同时，并不会带来太大的相噪恶化。

参考文献：

L. Kong and B. Razavi, “A 2.4 GHz 4 mW Integer-N Inductorless RF Synthesizer,” in IEEE Journal of Solid-State Circuits, vol. 51, no. 3, pp. 626-635, March 2016, doi: 10.1109/JSSC.2015.2511157.

以下的文献中也有用到这种滤波器结构， [2] Lee, Yongsun et al. “A Low-Jitter and Low-Reference-Spur Ring-VCO-Based Switched-Loop Filter PLL Using a Fast Phase-Error Correction Technique.” IEEE Journal of Solid-State Circuits 53 (2018): 1192-1202. 这篇文献介绍了一种带宽可调的TypeI PLL，利用数字Lock Detector进行判断调节，以达到减小锁定时间的同时减小带内相噪的效果。

优惠劵

Clara_D

关注

踩

觉得还不错?

一键收藏

知道了

P3 | msater-slave sampling filter (MSSF) (JSSC-2016-03)

今天看到一篇2016的JSSC，讲的是用Ring VCO的TypeI型PLL中滤波器结构的改进，从而带来了性能的提升。如下图所示，Fig.3是传统的滤波器结构，这种结构的Vctr的时域波形会有波动，给PLL带来spur。Fig.4是该文提出的新结构MSSF：msater-slave sampling filter，利用这种sample and hold的结构可以保持Vctr在环路锁定时输出值恒定。从而降低了杂散。该滤波器的传输函数如下：如果把这个滤波器当成零阶保持器(zero-order hold

复制链接

扫一扫

专栏目录

用于锁相环的两级ring VCO设计

12-12

用于锁相环的VCO设计，两级环路锁相环，该VCO输出频率范围宽

802.11协议详解

shanonhe的专栏

09-12

2574

WLAN协议详解

　　802.11b/g/n定义在2.4GHz频段中，802.11a/n/ac工作在5GHz频段中。

802.11：工作在2.4G频段，提供了每秒1兆或2兆的传输速率

802.11b：

　　* 最高11Mbps吞吐量

　　* 工作在2.4GHz，采用直序扩频（DSSS）

　　* 802.11b是所有无线局域网标准中最著名，也是普及最广的标准。在2.4GHz ISM频段中共有14个频宽为22MHz的频道可供使用，3个信道不重叠。

802.11g...

1 条评论

您还未登录，请先

后发表或查看评论

1x1 11b g n linux,基于RN1810下的2.4 GHz IEEE 802.11b/g/n无线模块

weixin_39645165的博客

05-13

221

特性• 符合IEEE 802.11b/g/n的收发器• 2.4 GHz IEEE 802.11n单流1x1• 与主机控制器的UART接口(4线，包括RTS/CTS)• 易于集成到最终产品中——最大程度地减少产品开发工作量，缩短上市时间• 使用简单的ASCII命令进行配置• 带稳压电路、晶振、RF匹配电路、功率放大器(Power Amplifier，PA)、低噪声放大器(LowNoise Ampli...

Redis主从复制和哨兵模式

hebeind100的博客

09-12

252

Redis主从复制

1.Master可以拥有多个slave

2.多个slave可以连接同一个Master外，还可以连接到其他的slave

3.主从复制不会阻塞Master在主从复制时，Master可以处理client请求。

4.提供系统的伸缩性。

主从复制的过程

1.slave与Master建立连接，发送sync同步命令。

也就是说当用户在Master写入一条命令后，他们之间会通过...

2.4GHz RF-SIM卡及读写终端技术概述

junllee的专栏

06-09

4253

2.4GHz RF-SIM卡及读写终端技术概述 1. 2.4GHz RF-SIM卡技术概要 (1)、使用2.4GHz频段, 自动选频； (2)、通信速率1Mbps，高可靠性连接与通信； (3)、支持自动感应和主动触发连接两种通信方法； (4)、双向通信距离10CM-500CM，可以根据应用调整； (5)、单向数据广播（半径100M）； (6

PLL Performance，Simulation and Design 4th学习笔记——Chapter1

张硕的博客

05-29

1101

Chapter1：基本PLL概述

VCO：Voltage-Controlled Oscillator压控振荡器，对于给定的一个输入电压，输出特定频率的振荡信号。

PLL：Phase-Locked Loops锁相环，输出特定频率，特定相位的振荡信号。

基本PLL操作和术语

基本模块

输入

输出

传输函数

说明

XTAL

参考频率ϕref\phi_{ref}ϕref

稳定的晶振提供输入参考频率（这里用相位，便于理解鉴相器）

R Counter

ϕref\phi_{ref

Ring VCO设计

努力学习的IC小白

07-21

423

Ring VCO设计

高速serdes技术学习总结

lxm920714的专栏

10-11

6029

Channel的特性

channel的特性。这里第一张图给出了三种它的传输函数，或者说它的s21。在不同的长度，loss区别很大。我们看到第一个蓝色的，它的loss就会比较小，在2.5Gbps时loss大概不到5db;而红色的这种比较长，它的loss在2.5Gbps的时候就有10个dB多一些。这个绿色的一根线，它不光loss多一点，还在10Gbps时有一个很差的一个点的loss，到-60多dB。

一般在设计的时候，会拿一个channel的model一般是Sparameter或者是LGC这种model，然后

【学习笔记/PLL】锁相环PLL线性模型理论分析

m0_57592021的博客

06-18

7088

本文参考了拉扎维、Sam palermo等学者的教材、文章，简略的总结了PLL线性模型下的噪声传递函数，欢迎各位同学批评指正。

P10 | Saturated PFD technique (TCAS I-2018-01)

Clara_D的博客

06-28

318

这篇文章做的是2.4GHz频段的Type I PLL，利用gain-boosted saturated PFD扩展了PLL的锁定范围，同时利用S&H电路的synchronous peak tracking loop filter降低了I类PLL的参考杂散。

提出的PLL框架如下：

什么叫saturated PFD？

如下图所示，传统PFD会存在一个cycle slipping的问题。

Conventional 3-state PFDs suffer from cycle-slipping, at

高性能全差分双环路VCO的设计

02-24

设计了一种基于SMIC0.18μm射频1P6MCMOSCraft.io的高性能全差分环形压控振荡器（ring-VCO），采用双环连接方式，并利用分立正反馈来提高性能。在1.8V电源电压下对电路进行仿真，结果表明：1）中心频率为500MHz的环形VCO频率调谐范围为341〜658MHz，增益误差Kvco为-278.8MHz / V，谐振在500MHz下VCO的幅度噪声为-104dBc / Hz @ 1MHz，功率为22mW; 2）中心频率为2.5GHz的环形VCO频率调谐范围为2.27〜2.79GHz，增益灵敏度Kvco为-514.6MHz / V，谐振在2.5GHz下VCO的振幅噪声为-98dBc / Hz @ 1MHz，功耗为32mW。该VCO适用于低压电路，高精度锁相环等。

一种基于Ring-VCO结构的宽频带低抖动锁相环的设计与实现

10-15

为了在高速传输系统中实现宽频带和低抖动时钟输出的要求，设计了一种基于Ring-VCO结构的低抖动锁相环，采用与锁相环锁定频率强相关的环路带宽调整方法来降低环路噪声，加速环路锁定，即利用全局参考调节电路中比较器模块将锁定控制电压与参考电压比较来改变各模块电流，根据不同锁定频率调整环路参数，大大缩短了锁定时间，同时利用四级差分环形振荡器和占空比调整电路的差分对称结构，降低了电路噪声。电路采用40 nm CMOS工艺实现，测试结果表明输出频率为1.062 5 GHz~5 GHz，在最高时钟频率5 GHz下眼图质量良好，时钟抖动39.6 ps。

PyPI 官网下载 | glb-slave-0.0.17.tar.gz

01-12

资源来自pypi官网。资源全名：glb-slave-0.0.17.tar.gz

nsmysql-slave架构图

Page restricted | ScienceDirect

Your Browser is out of date.

Update your browser to view ScienceDirect.

View recommended browsers.

Request details:

Request ID: 860a6fdbbc650974-HKG

IP: 49.157.13.121

UTC time: 2024-03-07T12:02:30+00:00

Browser: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/97.0.4692.71 Safari/537.36

About ScienceDirect

Shopping cart

Contact and support

Terms and conditions

Cookies are used by this site. By continuing you agree to the use of cookies.

Copyright Â© 2024 Elsevier B.V., its licensors, and contributors. All rights are reserved, including those for text and data mining, AI training, and similar technologies. For all open access content, the Creative Commons licensing terms apply.

Remote Sensing | Free Full-Text | MSSF: A Novel Mutual Structure Shift Feature for Removing Incorrect Keypoint Correspondences between Images

Next Article in Journal

A Multi-Scale Edge Constraint Network for the Fine Extraction of Buildings from Remote Sensing Images

Next Article in Special Issue

Geometric Prior-Guided Self-Supervised Learning for Multi-View Stereo

Previous Article in Journal

A Comprehensive Database of Indonesian Dams and Its Spatial Distribution

Previous Article in Special Issue

Multiscale Feature Fusion for the Multistage Denoising of Airborne Single Photon LiDAR

Journals

Active Journals

Find a Journal

Proceedings Series

Topics

Information

For Authors

For Reviewers

For Editors

For Librarians

For Publishers

For Societies

For Conference Organizers

Open Access Policy

Institutional Open Access Program

Special Issues Guidelines

Editorial Process

Research and Publication Ethics

Article Processing Charges

Awards

Testimonials

Author Services

Initiatives

Sciforum

MDPI Books

Preprints.org

Scilit

SciProfiles

Encyclopedia

JAMS

Proceedings Series

About

Overview

Contact

Careers

News

Press

Blog

Notice

You can make submissions to other journals

here.

clear

Notice

You are accessing a machine-readable page. In order to be human-readable, please install an RSS reader.

Continue

Cancel

clear

All articles published by MDPI are made immediately available worldwide under an open access license. No special

permission is required to reuse all or part of the article published by MDPI, including figures and tables. For

articles published under an open access Creative Common CC BY license, any part of the article may be reused without

permission provided that the original article is clearly cited. For more information, please refer to

https://www.mdpi.com/openaccess.

Feature papers represent the most advanced research with significant potential for high impact in the field. A Feature

Paper should be a substantial original Article that involves several techniques or approaches, provides an outlook for

future research directions and describes possible research applications.

Feature papers are submitted upon individual invitation or recommendation by the scientific editors and must receive

positive feedback from the reviewers.

Editor’s Choice articles are based on recommendations by the scientific editors of MDPI journals from around the world.

Editors select a small number of articles recently published in the journal that they believe will be particularly

interesting to readers, or important in the respective research area. The aim is to provide a snapshot of some of the

most exciting work published in the various research areas of the journal.

Original Submission Date Received: .

You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.

clear

zoom_out_map

Journals

Active Journals

Find a Journal

Proceedings Series

Topics

Information

For Authors

For Reviewers

For Editors

For Librarians

For Publishers

For Societies

For Conference Organizers

Open Access Policy

Institutional Open Access Program

Special Issues Guidelines

Editorial Process

Research and Publication Ethics

Article Processing Charges

Awards

Testimonials

Author Services

Initiatives

Sciforum

MDPI Books

Preprints.org

Scilit

SciProfiles

Encyclopedia

JAMS

Proceedings Series

About

Overview

Contact

Careers

News

Press

Blog

Submit

Search for Articles:

Title / Keyword

Author / Affiliation / Email

Journal

All Journals

Acoustics

Acta Microbiologica Hellenica

Actuators

Administrative Sciences

Adolescents

Advances in Respiratory Medicine (ARM)

Aerobiology

Aerospace

Agriculture

AgriEngineering

Agrochemicals

Agronomy

Air

Algorithms

Allergies

Alloys

Analytica

Analytics

Anatomia

Anesthesia Research

Animals

Antibiotics

Antibodies

Antioxidants

Applied Biosciences

Applied Mechanics

Applied Microbiology

Applied Nano

Applied Sciences

Applied System Innovation (ASI)

AppliedChem

AppliedMath

Aquaculture Journal

Architecture

Arthropoda

Arts

Astronomy

Atmosphere

Atoms

Audiology Research

Automation

Axioms

Bacteria

Batteries

Behavioral Sciences

Beverages

Big Data and Cognitive Computing (BDCC)

BioChem

Bioengineering

Biologics

Biology

Biology and Life Sciences Forum

Biomass

Biomechanics

BioMed

Biomedicines

BioMedInformatics

Biomimetics

Biomolecules

Biophysica

Biosensors

BioTech

Birds

Blockchains

Brain Sciences

Buildings

Businesses

Cancers

Cardiogenetics

Catalysts

Cells

Ceramics

Challenges

ChemEngineering

Chemistry

Chemistry Proceedings

Chemosensors

Children

Chips

CivilEng

Clean Technologies (Clean Technol.)

Climate

Clinical and Translational Neuroscience (CTN)

Clinics and Practice

Clocks & Sleep

Coasts

Coatings

Colloids and Interfaces

Colorants

Commodities

Complications

Compounds

Computation

Computer Sciences & Mathematics Forum

Computers

Condensed Matter

Conservation

Construction Materials

Corrosion and Materials Degradation (CMD)

Cosmetics

COVID

Crops

Cryptography

Crystals

Current Issues in Molecular Biology (CIMB)

Current Oncology

Dairy

Data

Dentistry Journal

Dermato

Dermatopathology

Designs

Diabetology

Diagnostics

Dietetics

Digital

Disabilities

Diseases

Diversity

DNA

Drones

Drugs and Drug Candidates (DDC)

Dynamics

Earth

Ecologies

Econometrics

Economies

Education Sciences

Electricity

Electrochem

Electronic Materials

Electronics

Emergency Care and Medicine

Encyclopedia

Endocrines

Energies

Eng

Engineering Proceedings

Entropy

Environmental Sciences Proceedings

Environments

Epidemiologia

Epigenomes

European Burn Journal (EBJ)

European Journal of Investigation in Health, Psychology and Education (EJIHPE)

Fermentation

Fibers

FinTech

Fire

Fishes

Fluids

Foods

Forecasting

Forensic Sciences

Forests

Fossil Studies

Foundations

Fractal and Fractional (Fractal Fract)

Fuels

Future

Future Internet

Future Pharmacology

Future Transportation

Galaxies

Games

Gases

Gastroenterology Insights

Gastrointestinal Disorders

Gastronomy

Gels

Genealogy

Genes

Geographies

GeoHazards

Geomatics

Geosciences

Geotechnics

Geriatrics

Gout, Urate, and Crystal Deposition Disease (GUCDD)

Grasses

Hardware

Healthcare

Hearts

Hemato

Hematology Reports

Heritage

Histories

Horticulturae

Hospitals

Humanities

Humans

Hydrobiology

Hydrogen

Hydrology

Hygiene

Immuno

Infectious Disease Reports

Informatics

Information

Infrastructures

Inorganics

Insects

Instruments

International Journal of Environmental Research and Public Health (IJERPH)

International Journal of Financial Studies (IJFS)

International Journal of Molecular Sciences (IJMS)

International Journal of Neonatal Screening (IJNS)

International Journal of Plant Biology (IJPB)

International Journal of Translational Medicine (IJTM)

International Journal of Turbomachinery, Propulsion and Power (IJTPP)

International Medical Education (IME)

Inventions

IoT

ISPRS International Journal of Geo-Information (IJGI)

Journal of Ageing and Longevity (JAL)

Journal of Cardiovascular Development and Disease (JCDD)

Journal of Clinical & Translational Ophthalmology (JCTO)

Journal of Clinical Medicine (JCM)

Journal of Composites Science (J. Compos. Sci.)

Journal of Cybersecurity and Privacy (JCP)

Journal of Developmental Biology (JDB)

Journal of Experimental and Theoretical Analyses (JETA)

Journal of Functional Biomaterials (JFB)

Journal of Functional Morphology and Kinesiology (JFMK)

Journal of Fungi (JoF)

Journal of Imaging (J. Imaging)

Journal of Intelligence (J. Intell.)

Journal of Low Power Electronics and Applications (JLPEA)

Journal of Manufacturing and Materials Processing (JMMP)

Journal of Marine Science and Engineering (JMSE)

Journal of Market Access & Health Policy (JMAHP)

Journal of Molecular Pathology (JMP)

Journal of Nanotheranostics (JNT)

Journal of Nuclear Engineering (JNE)

Journal of Otorhinolaryngology, Hearing and Balance Medicine (JOHBM)

Journal of Personalized Medicine (JPM)

Journal of Pharmaceutical and BioTech Industry (JPBI)

Journal of Respiration (JoR)

Journal of Risk and Financial Management (JRFM)

Journal of Sensor and Actuator Networks (JSAN)

Journal of Theoretical and Applied Electronic Commerce Research (JTAER)

Journal of Vascular Diseases (JVD)

Journal of Xenobiotics (JoX)

Journal of Zoological and Botanical Gardens (JZBG)

Journalism and Media

Kidney and Dialysis

Kinases and Phosphatases

Knowledge

Laboratories

Land

Languages

Laws

Life

Limnological Review

Lipidology

Liquids

Literature

Livers

Logics

Logistics

Lubricants

Lymphatics

Machine Learning and Knowledge Extraction (MAKE)

Machines

Macromol

Magnetism

Magnetochemistry

Marine Drugs

Materials

Materials Proceedings

Mathematical and Computational Applications (MCA)

Mathematics

Medical Sciences

Medical Sciences Forum

Medicina

Medicines

Membranes

Merits

Metabolites

Metals

Meteorology

Methane

Methods and Protocols (MPs)

Metrology

Micro

Microbiology Research

Micromachines

Microorganisms

Microplastics

Minerals

Mining

Modelling

Molbank

Molecules

Multimodal Technologies and Interaction (MTI)

Muscles

Nanoenergy Advances

Nanomanufacturing

Nanomaterials

NDT

Network

Neuroglia

Neurology International

NeuroSci

Nitrogen

Non-Coding RNA (ncRNA)

Nursing Reports

Nutraceuticals

Nutrients

Obesities

Oceans

Onco

Optics

Oral

Organics

Organoids

Osteology

Oxygen

Parasitologia

Particles

Pathogens

Pathophysiology

Pediatric Reports

Pharmaceuticals

Pharmaceutics

Pharmacoepidemiology

Pharmacy

Philosophies

Photochem

Photonics

Phycology

Physchem

Physical Sciences Forum

Physics

Physiologia

Plants

Plasma

Platforms

Pollutants

Polymers

Polysaccharides

Poultry

Powders

Proceedings

Processes

Prosthesis

Proteomes

Psych

Psychiatry International

Psychoactives

Publications

Quantum Beam Science (QuBS)

Quantum Reports

Quaternary

Radiation

Reactions

Real Estate

Receptors

Recycling

Religions

Remote Sensing

Reports

Reproductive Medicine (Reprod. Med.)

Resources

Rheumato

Risks

Robotics

Ruminants

Safety

Sci

Scientia Pharmaceutica (Sci. Pharm.)

Sclerosis

Seeds

Sensors

Separations

Sexes

Signals

Sinusitis

Smart Cities

Social Sciences

Société Internationale d’Urologie Journal (SIUJ)

Societies

Software

Soil Systems

Solar

Solids

Spectroscopy Journal

Sports

Standards

Stats

Stresses

Surfaces

Surgeries

Surgical Techniques Development

Sustainability

Sustainable Chemistry

Symmetry

SynBio

Systems

Targets

Taxonomy

Technologies

Telecom

Textiles

Thalassemia Reports

Thermo

Tomography

Tourism and Hospitality

Toxics

Toxins

Transplantology

Trauma Care

Trends in Higher Education

Tropical Medicine and Infectious Disease (TropicalMed)

Universe

Urban Science

Uro

Vaccines

Vehicles

Venereology

Veterinary Sciences

Vibration

Virtual Worlds

Viruses

Vision

Waste

Water

Wind

Women

World

World Electric Vehicle Journal (WEVJ)

Youth

Zoonotic Diseases

Article Type

All Article Types

Article

Review

Communication

Editorial

Abstract

Book Review

Brief Report

Case Report

Comment

Commentary

Concept Paper

Conference Report

Correction

Creative

Data Descriptor

Discussion

Entry

Essay

Expression of Concern

Extended Abstract

Guidelines

Hypothesis

Interesting Images

Letter

New Book Received

Obituary

Opinion

Perspective

Proceeding Paper

Project Report

Protocol

Registered Report

Retraction

Short Note

Study Protocol

Systematic Review

Technical Note

Tutorial

Viewpoint

Advanced Search

Section

Special Issue

Volume

Issue

Number

Page

Logical OperatorOperator

AND

Search Text

Search Type

All fields

Title

Abstract

Keywords

Authors

Affiliations

Doi

Full Text

References

add_circle_outline

remove_circle_outline

Journals

Remote Sensing

Volume 15

Issue 4

10.3390/rs15040926

Submit to this Journal

Review for this Journal

Propose a Special Issue

►

▼

Article Menu

Academic Editor

Riccardo Roncella

Subscribe SciFeed

More by Authors Links

on DOAJ

Liu, J.

Sun, K.

Jiang, S.

Li, K.

Tao, W.

on Google Scholar

Liu, J.

Sun, K.

Jiang, S.

Li, K.

Tao, W.

on PubMed

Liu, J.

Sun, K.

Jiang, S.

Li, K.

Tao, W.

/ajax/scifeed/subscribe

Article Views

Citations

Table of Contents

Altmetric

announcement

Help

format_quote

Cite

question_answer

Discuss in SciProfiles

thumb_up

...

Endorse

textsms

...

Comment

Need Help?

Support

Find support for a specific problem in the support section of our website.

Get Support

Feedback

Please let us know what you think of our products and services.

Give Feedback

Information

Visit our dedicated information section to learn more about MDPI.

Get Information

clear

JSmol Viewer

clear

first_page

Download PDF

settings

Order Article Reprints

Font Type:

Arial

Georgia

Verdana

Font Size:

Line Spacing:



Column Width:



Background:

Open AccessArticle

MSSF: A Novel Mutual Structure Shift Feature for Removing Incorrect Keypoint Correspondences between Images

Juan LiuJuan Liu

Scilit

Preprints.org

Google Scholar

View Publications

1,2, Kun SunKun Sun

Scilit

Preprints.org

Google Scholar

View Publications

1,2,*, San JiangSan Jiang

Scilit

Preprints.org

Google Scholar

View Publications

1, Kunqian LiKunqian Li

Scilit

Preprints.org

Google Scholar

View Publications

3 and Wenbing TaoWenbing Tao

Scilit

Preprints.org

Google Scholar

View Publications

2,4

Hubei Key Laboratory of Intelligent Geo-Information Processing, School of Computer Sciences, China University of Geosciences, Wuhan 430074, China

Key Laboratory of Image Processing and Intelligent Control (Huazhong University of Science and Technology), Ministry of Education, Wuhan 430074, China

College of Engineering, Ocean University of China, Qingdao 266100, China

National Key Laboratory of Science and Technology on Multi-Spectral Information Processing, School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan 430074, China

Author to whom correspondence should be addressed.

Remote Sens. 2023, 15(4), 926; https://doi.org/10.3390/rs15040926

Submission received: 15 November 2022

Revised: 1 February 2023

Accepted: 2 February 2023

Published: 8 February 2023

(This article belongs to the Special Issue Machine Learning for LiDAR Point Cloud Analysis)

Download keyboard_arrow_down

Download PDF

Download PDF with Cover

Download XML

Download Epub

Browse Figures

Versions Notes

Abstract:

Removing incorrect keypoint correspondences between two images is a fundamental yet challenging task in computer vision. A popular pipeline first computes a feature vector for each correspondence and then trains a binary classifier using these features. In this paper, we propose a novel robust feature to better fulfill the above task. The basic observation is that the relative order of neighboring points around a correct match should be consistent from one view to another, while it may change a lot for an incorrect match. To this end, the feature is designed to measure the bidirectional relative ranking difference for the neighbors of a reference correspondence. To reduce the negative effect of incorrect correspondences in the neighborhood when computing the feature, we propose to combine spatially nearest neighbors with geometrically “good” neighbors. We also design an iterative neighbor weighting strategy, which considers both goodness and correctness of a correspondence, to enhance correct correspondences and suppress incorrect correspondences. As the relative order of neighbors encodes structure information between them, we name the proposed feature the Mutual Structure Shift Feature (MSSF). Finally, we use the proposed features to train a random forest classifier in a supervised manner. Extensive experiments on both raw matching quality and downstream tasks are conducted to verify the performance of the proposed method.

Keywords: structure shift feature; mismatch removal; image matching; 3D reconstruction; pose estimation

Graphical Abstract

1. IntroductionMatching feature points between two images are widely used in many computer vision tasks [1,2,3,4,5,6,7]. Since SIFT [8] achieved great success two decades ago, the descriptor-based method has become more and more popular. Given detected keypoints, a lot of handcrafted [9,10,11] or learned [12,13,14,15,16] descriptors were proposed to search for reliable correspondences between two views. However, due to challenges, such as large geometric distortion, partial overlapping and local ambiguity, the initial matches might be contaminated by a high ratio of incorrect correspondences. To alleviate this problem, a mismatch removal method is usually applied as a post-processing step.While incorrect matching points exhibit ambiguity in the feature space, they have quite different geometric or spatial properties from correct ones. Based on this observation, existing methods perform in an unsupervised or supervised manner. Unsupervised methods impose global constraints such as global epipolar geometry [17] and motion coherency [18,19,20]. Some of these methods impose semi-global constraints, such as piecewise smooth transformation [21,22] and local graph structure [23,24,25], on the tentative matches. The idea of using local structure or geometry information has also been explored in other articles [26,27,28]. The corresponding pairs that violate these conditions will be rejected as outliers. Such kinds of methods dig deeper into the local strcuture of the matching points and prove to work well. However, outliers contained in the local neighborhood destroy the original structure, which poses new challenges to these methods. By contrast, supervised methods treat mismatch removal as a classification problem. In such kinds of methods, each matching pair is associated with a feature vector computed by handcrafted rules or deep neural networks. Then, a classifier is learned in the training stage and then predicts whether a putative correspondence is positive or negative in the testing phase [29,30,31]. Nevertheless, how to design proper features still remains a non-trivial task.In this paper, we propose a new feature for each putative correspondence and use it in a learning-based method to identify incorrect matches. The observation is that for a correct match, the relative order for several of its neighbors should be stable from one view to another. By contrast, the difference between the relative order for the neighbors of an incorrect match will be obviously large. Following this idea, a feature vector representing the relative order difference for the neighbors of a reference correspondence is computed. However, such a feature is dependent on the direction of two images. That is, the features will be different when computed forward and backward. To remove the asymmetry, we first compute the feature vector from the first image to the second image and do the same thing in a reverse direction. Finally, we concatenate them to obtain a higher dimensional feature vector. Since this feature vector encodes the bidirectional local structure shift of a putative correspondence on both views, we name it the Mutual Structure Shift Feature (MSSF). Ideally, correct matches can better preserve the local structure, so they present a small ranking difference and tend to distribute near the origin of the MSSF space. In contrast, wrong matches will spread far away from the origin. In this way, inliers and outliers can be distinguished more clearly.Another issue we are facing is how to define the local neighborhood when computing the proposed MSSF. Using spatially nearest neighbors is intuitive, but outliers might inevitably be involved. In this case, the feature of a correct match will shift towards the domain of incorrect matches along certain dimensions, making classification more difficult. A toy example is given in Figure 1. Figure 1a visualizes our MSSF by mapping it to a lower dimensional space when the neighborhood is contaminated by outliers. As we can see, there is significant overlap between positive and negative samples, making classification harder, while Figure 1b is the ideal case when the neighborhood contains no outliers. Compared with Figure 1a, the distributions of positive and negative samples are more compact and discriminative. The number of points mixed with a different class is also reduced. Inspired by the above observation, we make two improvements to our algorithm. First, we use geometrically “good” neighbors in conjunction with spatially nearest neighbors to reduce the risk of involving outliers. Second, we design an iterative weighting approach to enhance inliers and suppress outliers. Specifically, in each step, each neighboring correspondence is weighted by two scores: goodness and correctness. Goodness indicates whether a neighboring correspondence shares similar geometric properties with the target, and correctness reflects the confidence of a neighboring correspondence predicted by our model. As the iteration goes on, the weights of suspected mismatches gradually shrink, and the MSSF will be less affected by outliers. Finally, a random forest classifier trained with the proposed MSSF is used to distinguish correct matches from incorrect matches.Briefly, the contribution of this paper lies in the following aspects. (1) We propose a novel Mutual Structure Shift Feature (MSSF) to better distinguish correct and incorrect matches. The main observation is that the neighbors of correct matches are more likely to have consistent rankings across different views. (2) We propose to combine geometrically “good” neighbors with spatially nearest neighbors, which reduces the risk of involving outliers. (3) We propose an iterative weighting strategy considering both goodness and correctness of a match to enhance inliers and suppress outliers. As a result, our feature shows good distribution property and is more friendly to the classification task. 2. Related Work 2.1. Traditional MethodsAs a hot topic in computer vision, mismatch removal has been well studied in the past few years. Being one of the most famous robust model estimators, RANSAC and its variants [32,33,34] have been widely used in many modern applications such as SfM and SLAM. It estimates a parametric two-view geometry model in a re-sampling fashion, and removes correspondences with too large fitting errors. Different from an explicit binocular geometric model, ICF [35] learns two matching functions to check the consistency of putative matches in which each matching function regresses the position of a matching point from one image to another. However, the relationship between correspondences is somewhat ignored. In VFC [19] and its variants [18,20,36], correspondences are supposed to agree with a non-rigid motion function in a Bayesian framework. An additional regularization term is introduced to impose smoothness and coherence constraint. The unique global coherent restriction is extended by CODE [37] in which the non-linear regression formulation accommodates different local motion types with spatial discontinuities. Recently, the RFM [38] method has tried to find correct matches satisfying multiple local consistent motion patterns from a clustering view. The classical DBSCAN method is customized to achieve this goal.To capture local motion properties, a method called LPM [23] was proposed. By computing k nearest neighbors of a correct match on both images, the authors required that two neighborhood sets should have a large intersection. This problem is formulated as a convex optimization problem with a closed-form solution, which is more computationally efficient than the aforementioned iterative methods. Later, the intersection of the k-nn measurement in the LPM was replaced with the weighted Spearman’s footrule distance in mTopKRP [39]. This work revealed that rank information of neighbors has the potential for distinguishing correct from wrong matches. However, our method differs from it in two major differences. First, mTopKRP chooses k nearest points on two views separately. As a result, two sets of neighbors may not belong to the same matches, making similarity measuring intractable. Second, it does not consider the quality of k nearest neighbors, which may involve outliers. By contrast, the proposed method selects neighboring correspondences rather than merely points in two directions and designs a weighting strategy to suppress outliers. 2.2. Learning-Based MethodsApart from optimization-based methods, some researchers resort to learning algorithms to identify incorrect matches, which is essentially binary classification. Ma et al. [30] proposed a handcrafted feature for the classification task. Their combinatorial 33-dimensional feature fuses different attributes of a local neighborhood, such as percentage of intersection, ratio of length and angle. Moo Yi et al. [29] proposed the first work using deep learning, LGC. Inspired by the successful experience in point cloud processing, they designed an architecture based on Multi-Layer Perceptrons and ResNet blocks to extract features from each correspondence. Their training aims to minimize both classification loss and geometric loss. However, neighborhood information was not considered when extracting features. In a more recent work, NM-Net [31], the authors defined a graph around each correspondence and performed feature extraction with a graph convolution network. To avoid involving outliers in the graph, a good neighbor mining strategy was designed. Only classification loss was minimized because the structure constraint has already been integrated in the graph representation. This work was improved by CLNet [40], which progressively learns local-to-global consensus on dynamic graphs to prune outliers. To explore the complex context of putative correspondences, OANet [41] introduced a DiffPool layer and an Order-Aware DiffUnpool layer to capture local context. Moreover, it also developed order-aware filtering blocks to capture the global context. A novel smooth function which fits coherent motions on a graph of correspondences is proposed in LMCNet [42]. Based on its closed-form solution, a differentiable layer is designed in a deep neural network. ACN [43] is a simple yet effective technique to build permutation-equivariant networks. It normalizes the feature maps with weights estimated within the network, which can effectively exclude outliers.Different from the above methods which rely on complicated principles or large networks, in this paper we propose a simpler yet effective feature to prune outliers, which measures the mutual structure shift between two views. 3. The Proposed MethodSuppose we have a pair of images

′

and an initial matching set

…

between them. Each match

consists of a pair of keypoints, i.e.,

′

, where

comes from I, and

′

comes from

′

. The position of a keypoint

is represented by its image coordinate

. Similarly,

′

. 3.1. The Mutual Structure Shift FeatureThe basic idea of the proposed feature is that for a correct match, the relative order of its neighbors should be stable from one view to another, while this does not apply to outliers. Here we simply use the spatial Euclidean distance between keypoints as the distance measurement. Specifically, we compute the spatial distances between all the remaining points and the reference point and then sort them in an ascending order. Then, we record the distance orders for the selected neighbors. Let us have a look at an example shown in Figure 2. In Figure 2a, the red line represents a wrong match, and the remaining are correct matches. For a certain reference match, the three rows of Figure 2b plot the distance rankings of its neighbors on the first image, the second image and their difference, respectively. From left to right, we will discuss three cases. (1) We refer to I as the reference match and use its five correct neighbors

{

}

. (2) The same as (1), but we replace one of the above five correct neighbors with an outlier O (the red line). The neighbors are now

{

}

. (3) We consider an incorrect match O (the red line) as the reference and its five correct neighbors

{

}

. As we can see from the last row of Figure 2b, if the reference correspondence and all of its neighbors are correct, the two rankings are similar, and the order difference is the smallest. If the reference correspondence is correct but its neighbors are contaminated by outliers, the order difference will significantly increase at the corresponding position. If the reference correspondence is an outlier, the two rankings are quite different, and the order difference is generally larger. In this example, the order difference can help us to distinguish inliers from outliers. Since it implicitly encodes the structure information around the reference correspondence, we name it as the structure shift feature.We denote the structure shift feature computed from I to

′

and that computed from

′

to I as

. Our mutual structure shift feature is a combination of

and

. Taking the former for an example, we first find neighbors of the reference correspondence on I and obtain the co-occurrence neighbors on

′

according to the correspondences. Denote

as the distance order vector of the neighbors on I and

′

as the distance order vector of the neighbors on

′

. Each element in the distance order vector is an integer in

[

−

]

, where n is the total number of correspondences. Then,

is defined as the absolute difference between the two order vectors, which can be expressed as:

−

′

(1)

Equation (1) measures the change in the order of neighbor points. If the values in

are closer to 0, the better structure is preserved. Similarly,

can be computed in the same way in a reverse direction (from

′

to I).Thus far, we have computed the structure shift feature for a reference correspondence

and

in both directions. Our mutual structure shift feature is then defined as:

⊕

(2)

where ⊕ is the concatenation operation. 3.2. Neighbor Selection: Nearest Neighbors and “Good” NeighborsAs our mutual structure shift feature is built upon the neighbors of a reference correspondence, we need to carefully consider the neighbor selection strategy. Suppose the number of neighbors is k. Using k nearest neighbors in the Euclidean space is intuitive and easy to implement. However, such neighbors are vulnerable to mismatches, and the quality of the mutual structure shift feature decays significantly when the ratio of inliers is very low.To address the above issue, we propose to adopt “good” neighbors instead of using spatially nearest neighbors only. In our context, “good” neighbors refer to those who share similar local geometric information with the reference correspondence. Specifically, we first use the Hessian detector to detects. The local

affine transformation matrices at a pair of matching keypoint

and

′

are denoted as A and

′

, respectively. Next, the geometric information matrix

and

′

for this correspondence are computed from the following equation:

′

(3)

Then, we calculate the

local homography transformation at each keypoint using the following equation.

′

−

′

−

…

(4)

where

maps

to a new position in

′

, and

′

maps

′

to a new position in I. We assume that “good” neighbors should have the same or similar local homography transformation with each other.Based on this assumption, we can compute the geometric consistency error between a pair of correspondences from:

(

•

)

−

(

•

)

(5)

′

(

′

•

′

)

−

(

′

•

′

)

(6)

where i and j are the indices of two correspondences,

converts homogeneous coordinates into non-homogeneous coordinates, and

returns the sum of absolute values of all the elements. Equation (5) indicates that if two keypoints

and

on image I are geometrically similar, the position after mapping

with its own homography transformation

should be close to the position after mapping it with

, which is the homography transformation of

. As a result, the geometric error

between keypoints

and

would be small. Similarly,

′

in Equation (6) reflects this property in a reverse direction. To regularize the errors to a particular interval, we compute a similarity score between any two correspondences by applying the following exponential mapping function:

−

…

(7)

′

−

′

…

(8)

Here

and

′

indicate the similarities between correspondences i and j from I to

′

and from

′

to I, respectively. Both

and

′

range between

[

]

. According to [31],

is a flexible parameter because it tunes the similarity values but does not change the ranking results. We set it to a constant

−

throughout this paper.Finally, the neighbors of a reference correspondence

(

′

)

consist of two kinds of neighbors, i.e., k nearest neighbors and k “good” neighbors. On the one hand, we find k nearest neighbors of both

and

′

on each image. On the other hand, we find the top k “good” neighbors on each image according to the similarity scores in Equation (7) and Equation (8). The mutual structure shift feature is computed from the union of these neighbors according to Equation (2). We can use these features to train a classifier to distinguish inliers from outliers. 3.3. Neighbor Weighting StrategyIn the previous neighbor selection stage, it is still difficult to avoid involving outliers. Hence, we design an iterative weighting strategy during testing to enhance inliers and suppress outliers in the neighborhood. At the very beginning, all the correspondences have the same weights so that they have equal chance to be chosen in the neighbor selection stage. In the following iterations, we first re-weight the correspondences given the prediction of the last iteration. Then, we compute new feature vectors based on the updated neighbors and feed them to the classifier. Please note that we only update features in each iteration, and the parameters of the random forest are fixed. The details of our iterative weighting strategy are as follows. When selecting nearest neighbors, we take the probability predicted by the classifier as the new confidence and select k nearest neighbors whose confidence is greater than a threshold. When selecting “good” neighbors, we use the following equation to re-weight the correspondences:

(

)

(

)

∗

(

)

…

(9)

′

(

)

′

(

)

∗

(

)

…

(10)

where ∗ is the multiplication operator,

(

)

and

(

)

are the similarity scores initialized by Equation (7) and updated after the

iteration, respectively, and

(

)

is the probability of the

correspondence predicted by our classifier after the

iteration. In the next iteration, neighbors are selected using the above updated similarity scores. We can see from Equations (9) and (10) that if an outlier is involved by mistake at the beginning, it could be removed in the following process as the confidence predicted by the classifier gradually reduces.Finally, a random forest classifier trained with the proposed MSSF feature is used to distinguish correct matches from incorrect matches. In our experiments, we use 40 decision trees in the forest. If the probability predicted by the classifier is greater than a threshold

, the correspondence is deemed correct. 4. Experiments 4.1. Datasets and SettingsFour public datasets were used in our experiments: DTU [44], DAISY [45], ChallengeData [46] and NMNET [31]. DTU is widely used for stereo matching. Images in each scene were taken at 49 or 64 different positions. The projection matrix of each view is provided as the ground truth. We used two recommended scenes scan1 and scan6 in our experiments. Each of them contained 180 image pairs. DAISY is a wide-baseline dataset which contains two scenes: fountain and herzjesu. There are 11 and 8 images in each scene, respectively. We created a total of 40 and 22 image pairs in each scene by matching adjacent images. Both the intrinsic and extrinsic parameters of each view are provided for evaluation. ChallengeData is designed for large scale Structure from Motion (SfM). Images in this dataset present different illumination, wide baseline and heavy occlusion, etc. The camera pose information reconstructed by a standard SfM pipeline [47] is provided as the ground truth. To surmount the excessive number of image pairs in this dataset, we selected three scenes: trevi_fountain, grand_place_brussels and hagia_sophia_interior for testing. Following the same protocol in [16], image pairs in each scene were classified into three categories according to the rotation angle: easy ([15

∘

, 30

∘

)), moderate ([30

∘

, 45

∘

)) and hard ([45

∘

, 60

∘

]). For each category, we randomly selected 100 image pairs for testing. Finally, we used the NMNet dataset to test our application on Unmanned Aerial Vehicle (UAV) images. This dataset was captured by a drone at four sites. For each site, there are two versions of data: wide baseline and narrow baseline. In our experiment, we used the more challenging wide baseline version and selected 10 image pairs from each site. To determine if a correspondence was correct, we checked if the Epipolar geometric distance error was below a threshold

. The default value of

was two.The proposed method was evaluated in two aspects. First, we employed precision (P), recall (R) and F1-score to see the raw matching quality. Moreover, we also report the F1-score, which was computed from:

∗

100

(11)

As we can see from Equation (11), the F1-score is a composite indicator of both precision and recall. Next, we also testes the accuracy of camera pose estimation, which is an important downstream task of feature correspondences. The performance was evaluated by the pose estimation accuracy. To be specific, we estimated the essential matrix between two views and recovered the rotation matrix R and translation vector t between them. Then, we measured the angle error by comparing the estimation with the ground truth using the following equations.

arccos

∗

−

∗

180

arccos

∗

180

(12)

In Equation (12), the subscript

and

represent the estimated value and the ground truth, respectively;

and

are the angle errors of the rotation matrix and the translation vector. In our experiments, the thresholds for both

and

were set to 10

∘

.We compared our approach with six mismatch elimination algorithms from recent years. These includes traditional methods, such as LPM [23], mTop [39], RANSAC [32] and RFM [38], and modern machine-learning-based methods, such as LMR [30] and LGC [29]. All the experiments were performed on a machine equipped with a Xeon E5-2680 CPU, 64GB RAM and a GeForce GTX 1080Ti GPU. 4.2. Parameter AnalysisThe size of neighborhood k is an important parameter in our method. It directly determines the dimension of the feature vector. On the one hand, smaller k not only limits the capacity of structure information in the feature, but also is sensitive to outliers and large deformation. On the other hand, larger k increases the risk of involving outliers in the neighborhood and will take up more resources as well. Hence, we analyzed k quantitatively by measuring the average F1-score and average running time on one of the scenes in ChallengeData. As shown in Figure 3, when k takes 4, 8, 16 and 32, the average F1-score first increases and then drops. Meanwhile, the average running time of each image pair keeps rising. To balance the two factors, we set k to 16 for all the experiments.Another parameter to be investigated is the threshold

of the predicted probability. Generally speaking, increasing

can eliminate more outliers (leading to higher precision) but may kill more correct matches by mistake (leading to lower recall). Similarly, using smaller

may result in lower precision and higher recall. To set a proper value for

, the distribution of the predicted probability for both correct and wrong correspondences were investigated. In the example shown in Figure 4, the number of correct and wrong correspondences are 28 k and 18 k, respectively. For more than

of the correct correspondences, their probabilities are greater than 0.7. We also note that the number of wrong correspondences whose probabilities are smaller than 0.4 accounts for nearly

of the total. This property is favored because two distributions have small overlap. In order to balance the matching accuracy between correct and wrong correspondences,

was set to 0.5 for all the experiments. 4.3. Ablation StudyDifferent kinds of neighbors. The quality of neighbors is important. In the proposed method, we use the combination of both nearest neighbors and “good” neighbors. We also test the results of using nearest neighbors or “good” neighbors only. Thus, we use the DTU dataset to test the three settings and report the average F1-score in Table 1. As we can see, using the nearest neighbors is easy to implement and depicts the local structure well. However, it results in the lowest F1-score. The main reason is that nearest neighbors are easily contaminated by outliers. Using “good” neighbors only will increase the F1-score, which verifies that fewer outliers are involved by considering structure compatibility. The best results are achieved by using both nearest neighbors and “good” neighbors, which is shown in the last row. In Figure 5, we visualize both nearest neighbors and “good” neighbors on two pairs of images in the DTU dataset. As we can see, “good” neighbors are not always spatially nearby samples but contain fewer incorrect correspondences. Figure 6 plots the average precision of two kinds of neighbors on scan1 and scan6 of the DTU dataset. We can see that the average precision of “good” neighbors is significantly higher than nearest neighbors.Mutual strategy. As stated before, measuring the structure shift is asymmetric with respect to the direction. That is, the results might be different when performing from I to

′

and from

′

to I. Hence, we adopt a mutual strategy which performs in both directions. Here we give the results with and without the mutual strategy in Table 2. Similar to Table 1, the average F1-scores on the DTU dataset are reported. It is worth noting that the mutual strategy will double the feature size and consume more resources. We can see that using a mutual strategy leads to better results. This shows that if it is hard to identify a mismatch in one direction, adopting the reverse direction makes up for it.Iterative neighbor weighting strategy. In this part, we investigate whether the proposed iterative neighbor weighting strategy can truly reduce the risk for neighbors being contaminated by outliers. Hence, we define the First Outlier Position (FOP) as the indicator. FOP is a positive integer which indicates the position where the first outlier appears in the neighbor sequence. In other words, neighbors ranking before FOP are all inliers. Figure 7 shows the average FOP after each iteration for all the correspondences on a pair of images. As we can see, at the very beginning, the first outlier on average appears at the 13.3-th position in the neighbor sequence. This value grows up to 19.8 and 23.7 for the second and third iteration, respectively. If we run more iterations, the growth slows down. This curve shows that our weighting strategy pushes wrong correspondences to the back of the neighbor sequence. Thus, when we select the top k neighbors, the risk of involving outliers is greatly reduced. If k is smaller than the FOP, our neighbors will contain no outliers. 4.4. Raw Matching Quality EvaluationIn order to verify the performance of our method, we calculate the average precision, recall and F1-score on seven scenarios from three datasets (two from DTU, three from ChallengeData and two from Daisy). For each scenario, we plot the cumulative distribution of precision, recall and F1-score for all image pairs in Figure 8. As we can see from this figure, the precision of our method is not always the best, but our recall is much higher than the other methods. As a result, our method has the best F1-score in most cases. This shows that our method can preserve correct correspondences as much as possible. We also note that for the first five scenes, there is not much difference between the F1-scores of different methods, except for RANSAC. However, it is even more significant for the last two scenes in Daisy. This shows that our method is superior to the other methods in generalization and stability.In Table 3, we calculate the average F1-score with different thresholds

for all the image pairs in Figure 8. As we can see, when

increases from 0.5 to 4, our method achieves the highest F1-score all the time. 4.5. Pose Estimation EvaluationHere, we evaluate the camera pose estimation accuracy of our method. We tested on two datasets: Daisy and ChallengeData. Daisy is a small-scale dataset, and the results are given in Table 4. As we can see, because of the very wide baseline in fountain, it is more challenging. and all the methods present lower accuracy than herzjesu. Our method has the highest accuracy for both translation and rotation. The gap between our method and the second best method is up to

. ChallengeData contains 300 image pairs in total, which is much larger than Daisy. The results on this dataset are given in Table 5. As we can see, when the difference between cameras increases (from easy to hard in each scene), the performance drops consistently. Similarly, our method obtains the best accuracy on both rotation and translation. 4.6. Application on UAV ImagesFinally, we evaluate the proposed method on UAV images. We first report the average F1-score in Table 6. The data show that our method improves the results better than the other methods. Next, we conducted relative pose estimation on this dataset and report the average angle error in Table 7. We can see that for scenes whose average F1-scores are higher than

, e.g., mao-wide and science-wide, the pose errors returned by all the methods could be no larger than

∘

. However, the pose errors for the compared methods easily rise to double digits for main-wide, whose average F1-scores are obviously lower than the above two scenes. Our method has the lowest errors for all the scenes. 5. ConclusionsIn this paper, we propose a new method to remove incorrect correspondences between two images. We found that for a correct reference correspondence, the distance rankings of its neighbors are consistent from one view to another. Based on this observation, we propose a new feature called the Mutual Structure Shift Feature (MSSF), which measures the bidirectional ranking difference for the neighbors of a reference correspondence. To compute MSSF, we combine both spatially nearest neighbors with geometrically consistent neighbors. In this way, the risk of involving outliers in the neighbors is effectively reduced. We also design an iterative weighting strategy to progressively enhance correct correspondences and suppress incorrect correspondences. Extensive experiments on both raw matching quality evaluation and downstream tasks are carried out, showing our method outperforms the other compared methods.In spite of the advantages, the limitations of our method lie in the following aspects. Firstly, since our method relies on information from the neighbors, its performance may deteriorate when we are not able to find enough qualified neighbors. This usually happens when the initial correspondences are too sparse or the inlier ratio is extremely low. Secondly, although our iterative weighting strategy can effectively exclude outliers in the neighbors, it cannot remove incorrect correspondences that are associated with high confidence by the model at the very beginning. That is, if an incorrect correspondence could not be clearly recognized in the early stage, it is harder to identify it later.

Author ContributionsConceptualization, J.L. and K.S.; methodology, K.S. and K.L.; software, J.L.; validation, J.L.; formal analysis, S.J., K.L. and W.T.; resources, S.J.; writing—original draft preparation, K.S.; writing—review and editing, K.S. and K.L.; visualization, J.L. and S.J.; supervision, K.S. and W.T. All authors have read and agreed to the published version of the manuscript.FundingThis work was supported by the National Natural Science Foundation of China No. 62176242 and No. 62176096, also in part by the National Natural Science Foundation of China No. 61906177 and No. 42001413.Data Availability StatementThe data presented in this study are openly available in [31,44,45,46].Conflicts of InterestThe authors declare no conflict of interest.ReferencesMa, J.; Jiang, X.; Fan, A.; Jiang, J.; Yan, J. Image Matching from Handcrafted to Deep Features: A Survey. Int. J. Comput. Vis. 2020, 129, 23–79. [Google Scholar] [CrossRef]Ma, X.; Xu, S.; Zhou, J.; Yang, Q.; Yang, Y.; Yang, K.; Ong, S.H. Point set registration with mixture framework and variational inference. Pattern Recognit. 2020, 104, 107345. [Google Scholar] [CrossRef]He, Q.; Zhou, J.; Xu, S.; Yang, Y.; Yu, R.; Liu, Y. Adaptive Hierarchical Probabilistic Model Using Structured Variational Inference for Point Set Registration. IEEE Trans. Fuzzy Syst. 2020, 28, 2784–2798. [Google Scholar] [CrossRef]Wang, T.; Jiang, Z.; Yan, J. Clustering-aware Multiple Graph Matching via Decayed Pairwise Matching Composition. In Proceedings of the Thirty-Fourth AAAI Conference on Artificial Intelligence, New York, NY, USA, 7–12 February 2020. [Google Scholar]Wang, R.; Yan, J.; Yang, X. Combinatorial Learning of Robust Deep Graph Matching: An Embedding based Approach. IEEE Trans. Pattern Anal. Mach. Intell. 2020. early access. [Google Scholar] [CrossRef]Min, J.; Lee, J.; Ponce, J.; Cho, M. Hyperpixel flow: Semantic correspondence with multi-layer neural features. In Proceedings of the IEEE International Conference on Computer Vision, Seoul, Republic of Korea, 27 October–2 November 2019; pp. 3395–3404. [Google Scholar]Tang, L.; Deng, Y.; Ma, Y.; Huang, J.; Ma, J. SuperFusion: A Versatile Image Registration and Fusion Network with Semantic Awareness. IEEE CAA J. Autom. Sin. 2022, 9, 2121–2137. [Google Scholar] [CrossRef]Lowe, D.G. Distinctive Image Features from Scale-Invariant Keypoints. Int. J. Comput. Vis. 2004, 60, 91–110. [Google Scholar] [CrossRef]Bay, H.; Tuytelaars, T.; Gool, L.V. SURF: Speeded Up Robust Features. In Proceedings of the European Conference on Computer Vision, Graz, Austria, 7–13 May 2006; Lecture Notes in Computer Science. Springer: Berlin/Heidelberg, Germany, 2006; Volume 3951, pp. 404–417. [Google Scholar] [CrossRef]Calonder, M.; Lepetit, V.; Strecha, C.; Fua, P. BRIEF: Binary Robust Independent Elementary Features. In Proceedings of the European Conference on Computer Vision, Heraklion, Greece, 5–11 September 2010; Lecture Notes in Computer Science. Springer: Berlin/Heidelberg, Germany, 2010; Volume 6314, pp. 778–792. [Google Scholar] [CrossRef]Leutenegger, S.; Chli, M.; Siegwart, R. BRISK: Binary Robust invariant scalable keypoints. In Proceedings of the IEEE International Conference on Computer Vision, Washington, DC, USA, 20–25 June 2011; pp. 2548–2555. [Google Scholar] [CrossRef]Tian, Y.; Fan, B.; Wu, F. L2-Net: Deep Learning of Discriminative Patch Descriptor in Euclidean Space. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, 21–26 July 2017; pp. 6128–6136. [Google Scholar] [CrossRef]Mishchuk, A.; Mishkin, D.; Radenovic, F.; Matas, J. Working hard to know your neighbor’s margins: Local descriptor learning loss. In Proceedings of the Annual Conference on Neural Information Processing Systems, Long Beach, CA, USA, 4–9 December 2017; pp. 4826–4837. [Google Scholar]Dusmanu, M.; Rocco, I.; Pajdla, T.; Pollefeys, M.; Sivic, J.; Torii, A.; Sattler, T. D2-Net: A Trainable CNN for Joint Description and Detection of Local Features. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Computer Vision Foundation, Long Beach, CA, USA, 15–20 June 2019; pp. 8092–8101. [Google Scholar] [CrossRef]Tian, Y.; Yu, X.; Fan, B.; Wu, F.; Heijnen, H.; Balntas, V. SOSNet: Second Order Similarity Regularization for Local Descriptor Learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2019), Long Beach, CA, USA, 16–20 June 2019; pp. 11016–11025. [Google Scholar]Wang, Q.; Zhou, X.; Hariharan, B.; Snavely, N. Learning Feature Descriptors Using Camera Pose Supervision. In Proceedings of the ECCV, Glasgow, UK, 23–28 August 2020; Springer: Berlin/Heidelberg, Germany, 2020; Volume 12346, pp. 757–774. [Google Scholar]Ranftl, R.; Koltun, V. Deep fundamental matrix estimation. In Proceedings of the European Conference on Computer Vision (ECCV), Munich, Germany, 8–14 September 2018; pp. 284–299. [Google Scholar]Ma, J.; Zhao, J.; Tian, J.; Bai, X.; Tu, Z. Regularized vector field learning with sparse approximation for mismatch removal. Pattern Recognit. 2013, 46, 3519–3532. [Google Scholar] [CrossRef]Ma, J.; Zhao, J.; Tian, J.; Yuille, A.L.; Tu, Z. Robust point matching via vector field consensus. IEEE Trans. Image Process. 2014, 23, 1706–1721. [Google Scholar] [CrossRef]Ma, J.; Wu, J.; Zhao, J.; Jiang, J.; Zhou, H.; Sheng, Q.Z. Nonrigid point set registration with robust transformation learning under manifold regularization. IEEE Trans. Neural Netw. Learn. Syst. 2018, 30, 3584–3597. [Google Scholar] [CrossRef]Liu, H.; Yan, S. Common visual pattern discovery via spatially coherent correspondences. In Proceedings of the 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, San Francisco, CA, USA, 13–18 June 2010; pp. 1609–1616. [Google Scholar]Lipman, Y.; Yagev, S.; Poranne, R.; Jacobs, D.W.; Basri, R. Feature matching with bounded distortion. ACM Trans. Graph. (TOG) 2014, 33, 1–14. [Google Scholar] [CrossRef]Ma, J.; Zhao, J.; Jiang, J.; Zhou, H.; Guo, X. Locality preserving matching. Int. J. Comput. Vis. 2019, 127, 512–531. [Google Scholar] [CrossRef]Bian, J.; Lin, W.Y.; Matsushita, Y.; Yeung, S.K.; Nguyen, T.D.; Cheng, M.M. Gms: Grid-based motion statistics for fast, ultra-robust feature correspondence. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, San Juan, PR, USA, 17–19 June 2017; pp. 4181–4190. [Google Scholar]Liu, H.; Zheng, C.; Li, D.; Zhang, Z.; Lin, K.; Shen, X.; Xiong, N.N.; Wang, J. Multi-perspective social recommendation method with graph representation learning. Neurocomputing 2022, 468, 469–481. [Google Scholar] [CrossRef]Lhuillier, M.; Quan, L. Image Interpolation by Joint View Triangulation. In Proceedings of the 1999 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Fort Collins, CO, USA, 23–25 June 1999; pp. 2139–2145. [Google Scholar]Lee, I.C.; He, S.; Lai, P.L.; Yilmaz, A. BUILDING Point Grouping Using View-Geometry Relations. In Proceedings of the ASPRS 2010 Annual Conference, San Diego, CA, USA, 26–30 April 2010. [Google Scholar]Takimoto, R.Y.; Challella das Neves, A.; de Castro Martins, T.; Takase, F.K.; de Sales Guerra Tsuzuki, M. Automatic Epipolar Geometry Recovery Using Two Images. IFAC Proc. Vol. 2011, 44, 3980–3985. [Google Scholar] [CrossRef]Moo Yi, K.; Trulls, E.; Ono, Y.; Lepetit, V.; Salzmann, M.; Fua, P. Learning to find good correspondences. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, 18–23 June 2018; pp. 2666–2674. [Google Scholar]Ma, J.; Jiang, X.; Jiang, J.; Zhao, J.; Guo, X. LMR: Learning a two-class classifier for mismatch removal. IEEE Trans. Image Process. 2019, 28, 4045–4059. [Google Scholar] [CrossRef]Zhao, C.; Cao, Z.; Li, C.; Li, X.; Yang, J. NM-Net: Mining reliable neighbors for robust feature correspondences. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Long Beach, CA, USA, 15–20 June 2019; pp. 215–224. [Google Scholar]Fischler, M.A.; Bolles, R.C. Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography. Commun. ACM 1981, 24, 381–395. [Google Scholar] [CrossRef]Chum, O.; Matas, J. Matching with PROSAC-progressive sample consensus. In Proceedings of the 2005 IEEE computer society conference on computer vision and pattern recognition (CVPR’05), San Diego, CA, USA, 20–26 June 2005; Volume 1, pp. 220–226. [Google Scholar]Tran, Q.H.; Chin, T.J.; Carneiro, G.; Brown, M.S.; Suter, D. In defence of RANSAC for outlier rejection in deformable registration. In Proceedings of the European Conference on Computer Vision, Florence, Italy, 7–13 October 2012; Springer: Berlin/Heidelberg, Germany, 2012; pp. 274–287. [Google Scholar]Li, X.; Hu, Z. Rejecting mismatches by correspondence function. Int. J. Comput. Vis. 2010, 89, 1–17. [Google Scholar] [CrossRef]Ma, J.; Zhou, H.; Zhao, J.; Gao, Y.; Jiang, J.; Tian, J. Robust feature matching for remote sensing image registration via locally linear transforming. IEEE Trans. Geosci. Remote Sens. 2015, 53, 6469–6481. [Google Scholar] [CrossRef]Lin, W.Y.; Wang, F.; Cheng, M.M.; Yeung, S.K.; Torr, P.H.; Do, M.N.; Lu, J. CODE: Coherence based decision boundaries for feature correspondence. IEEE Trans. Pattern Anal. Mach. Intell. 2017, 40, 34–47. [Google Scholar] [CrossRef]Jiang, X.; Ma, J.; Jiang, J.; Guo, X. Robust feature matching using spatial clustering with heavy outliers. IEEE Trans. Image Process. 2019, 29, 736–746. [Google Scholar] [CrossRef]Jiang, X.; Jiang, J.; Fan, A.; Wang, Z.; Ma, J. Multiscale locality and rank preservation for robust feature matching of remote sensing images. IEEE Trans. Geosci. Remote Sens. 2019, 57, 6462–6472. [Google Scholar] [CrossRef]Zhao, C.; Ge, Y.; Zhu, F.; Zhao, R.; Li, H.; Salzmann, M. Progressive Correspondence Pruning by Consensus Learning. In Proceedings of the ICCV, Montreal, BC, Canada, 11–17 October 2021; pp. 6444–6453. [Google Scholar]Zhang, J.; Sun, D.; Luo, Z.; Yao, A.; Chen, H.; Zhou, L.; Shen, T.; Chen, Y.; Quan, L.; Liao, H. OANet: Learning Two-View Correspondences and Geometry Using Order-Aware Network. IEEE Trans. Pattern Anal. Mach. Intell. 2022, 44, 3110–3122. [Google Scholar] [CrossRef]Liu, Y.; Liu, L.; Lin, C.; Dong, Z.; Wang, W. Learnable Motion Coherence for Correspondence Pruning. In Proceedings of the CVPR. Computer Vision Foundation, Virtual, 19–25 June 2021; pp. 3237–3246. [Google Scholar]Sun, W.; Jiang, W.; Trulls, E.; Tagliasacchi, A.; Yi, K.M. ACNe: Attentive Context Normalization for Robust Permutation-Equivariant Learning. In Proceedings of the CVPR. Computer Vision Foundation, Seattle, WA, USA, 14–19 June 2020; pp. 11283–11292. [Google Scholar]Aanæs, H.; Jensen, R.R.; Vogiatzis, G.; Tola, E.; Dahl, A.B. Large-Scale Data for Multiple-View Stereopsis. Int. J. Comput. Vis. 2016, 120, 153–168. [Google Scholar] [CrossRef]Tola, E.; Lepetit, V.; Fua, P. DAISY: An Efficient Dense Descriptor Applied to Wide-Baseline Stereo. IEEE Trans. Pattern Anal. Mach. Intell. 2010, 32, 815–830. [Google Scholar] [CrossRef]Jin, Y.; Mishkin, D.; Mishchuk, A.; Matas, J.; Fua, P.; Yi, K.M.; Trulls, E. Image Matching Across Wide Baselines: From Paper to Practice. Int. J. Comput. Vis. 2021, 129, 517–547. [Google Scholar] [CrossRef]Schönberger, J.L.; Frahm, J. Structure-from-Motion Revisited. In Proceedings of the CVPR, Las Vegas, NV, USA, 27–30 June 2016; pp. 4104–4113. [Google Scholar]

Figure 1.

A visualization of the proposed MSSF after mapping it to a lower dimensional space: (a) computed features when the neighborhood contains outliers; (b) computed features when the neighborhood contains no outliers. Compared with (a), red points and blue points in (b) distribute more compactly, and the number of points mixed with another class is also reduced. This property is more in agreement with the consequent classification task.

Figure 1.

Figure 2.

An example of the structure shift feature: (a) some correspondences on a pair of images; (b) from top to bottom: the distance rankings for the neighbors of a reference correspondence on the first image, the second image and their difference, respectively. From left to right: a correct reference correspondence (in yellow) with five correct neighbors (solid green lines), a correct reference correspondence (in yellow) with four correct neighbors (solid green lines) and a wrong neighbor point (solid red lines), a wrong reference correspondence (in red) with five correct neighbors (solid green lines).

Figure 2.

Figure 3.

The analysis of k on grand_place_brussels from ChallengeData: Left: the average F1-score; Right: the average running time for each image pair.

Figure 3.

The analysis of k on grand_place_brussels from ChallengeData: Left: the average F1-score; Right: the average running time for each image pair.

Figure 4.

The probability distribution for both correct (blue) and wrong (red) correspondences in grand_place_brussels from ChallengeData.

Figure 4.

The probability distribution for both correct (blue) and wrong (red) correspondences in grand_place_brussels from ChallengeData.

Figure 5.

Visualization of both nearest neighbors and “good” neighbors on two pairs of images in the DTU dataset. The top row is from scan1 and the bottom row is from scan6. The reference correspondence is in yellow. Correct and wrong correspondences are in green and red, respectively.

Figure 5.

Figure 6.

The average precision of both nearest neighbors and “good” neighbors on scan1 and scan2 in the DTU dataset.

Figure 6.

The average precision of both nearest neighbors and “good” neighbors on scan1 and scan2 in the DTU dataset.

Figure 7.

The average FOP after 4 iterations for all the correspondences on a pair of images.

Figure 7.

The average FOP after 4 iterations for all the correspondences on a pair of images.

Figure 8.

Results compared with several state-of-the-art methods. Each row is a scene. From top to bottom: scan1 and scan6 from DTU; temple_nara_japan, notre_dame_front_facade and taj_mahal from ChallengeData; fountain and herzjesu from Daisy. From left to right are: precision, recall and F1-score, respectively.

Figure 8.

Table 1.

Ablation study of the neighbor selection strategy. The average F1-score (%) on the DTU dataset for three settings: using nearest neighbors only, using “good” neighbors only and using both of them. The best results are in bold.

Table 1.

Nearest Neighbors“Good” NeighborsScan1Scan6✔-75.4380.23-✔78.1781.38✔✔78.9881.98

Table 2.

Ablation study of the mutual strategy. The average F1-score (%) on the DTU dataset for two settings; w/o Mutual: using features computed from I to

′

only; w/Mutual: using features computed from both I to

′

and

′

to I. The best results are in bold.

Table 2.

Ablation study of the mutual strategy. The average F1-score (%) on the DTU dataset for two settings; w/o Mutual: using features computed from I to

′

only; w/Mutual: using features computed from both I to

′

and

′

to I. The best results are in bold.

SceneScan1Scan6w/o Mutual79.2381.77w/ Mutual79.9882.13

Table 3.

The average F1-score (%) with different thresholds

for all the image pairs in Figure 8. The best results are in bold. The red numbers in the brackets indicate the improvement between the best and the second best results.

Table 3.

The average F1-score (%) with different thresholds

for all the image pairs in Figure 8. The best results are in bold. The red numbers in the brackets indicate the improvement between the best and the second best results.

ThresholdLMRLPMmTOPRFMRANSACLGCOurs0.582.6282.1081.8878.6571.9175.2383.52 (+0.9)184.5983.9083.7280.5472.3977.1886.01 (+1.42)1.585.3084.6184.4581.4272.3777.8386.97 (+1.67)286.0285.3685.2082.1772.2778.3887.86 (+1.84)2.586.8385.9985.9283.1771.9779.2488.83 (+2.0)387.3686.5086.3783.7171.6679.5589.34 (+1.98)3.587.6986.8586.7684.1771.3479.9689.75 (+2.06)487.8787.0786.9784.5071.0480.2390.07 (+2.2)

Table 4.

Relative pose estimation accuracy (%) on Daisy. Each column represents a scene. Each cell represents the accuracy of rotation estimation (left) and the accuracy of translation estimation (right). The estimation of an image pair is successful when the angle error is under a certain threshold (10

∘

). The best results are in bold. The red numbers in the brackets indicate the improvement between the best and the second best results.

Table 4.

∘

). The best results are in bold. The red numbers in the brackets indicate the improvement between the best and the second best results.

MethodFountainHerzjesuLMR72.50/70.0090.91/90.91LPM65.00/65.0086.36/86.36mTOP70.00/70.0086.36/86.36RFM67.50/67.5086.36/86.36RANSAC57.50/55.0077.27/86.36LGC57.50/55.0086.36/86.36Ours75.00/75.00(+2.5/+5.0)95.45/95.45(+4.54/+4.54)

Table 5.

Relativepose estimation accuracy (%) on three scene of ChallengeData. Each cell represents the accuracy of rotation estimation (left) and the accuracy of translation estimation (right). The estimation of an image pair is successful when the angle error is under a certain threshold (10

∘

). The best results are in bold. The red numbers in the brackets indicate the improvement between the best and the second best results.

Table 5.

∘

). The best results are in bold. The red numbers in the brackets indicate the improvement between the best and the second best results.

Methodtrevi_fountaingrand_place_brusselshagia_sophia_interiorEasyModerateHardEasyModerateHardEasyModerateHardLMR98.0/88.097.0/90.092.0/88.095.0/61.084.0/59.078.0/63.096.0/35.088.0/60.085.0/77.0LPM98.0/83.095.0/86.093.0/90.095.0/64.088.0/55.081.0/60.098.0/35.091.0/62.085.0/54.0mTOP98.0/89.095.0/87.093.0/87.093.0/61.084.0/49.078.0/55.098.0/35.091.0/60.086.0/75.0RFM98.0/85.087.0/81.090.0/83.093.0/56.081.0/52.076.0/64.096.0/35.089.0/59.084.0/76.0RANSAC96.0/81.087.0/74.082.0/80.084.0/34.072.0/42.064.0/46.087.0/27.079.0/50.055.0/56.0LGC92.0/77.078.0/68.076.0/73.090.0/49.076.0/47.064.0/48.094.0/30.087.0/49.064.0/56.0Ours100.0/92.0 (+2.0/+3.0)99.0/92.0 (+2.0/+2.0)95.0/91.0 (+2.0/+1.0)96.0/66.0 (+1.0/+2.0)91.0/63.0 (+3.0/+4.0)84.0/73.0 (+3.0/+9.0)99.0/39.0 (+1.0/+4.0)95.0/64.0 (+4.0/+2.0)90.0/82.0 (+4.0/+5.0)

Table 6.

The average F1-score (%) of four scenes from the NMNET dataset. The best results are in bold. The red numbers in the brackets indicate the improvement between the best and the second best results.

Table 6.

The average F1-score (%) of four scenes from the NMNET dataset. The best results are in bold. The red numbers in the brackets indicate the improvement between the best and the second best results.

Lib-WideMain-WideMao-WideScience-WideLMR82.1174.4794.5891.44LPM86.0876.5195.3693.09mTOP85.4187.9095.2491.11RFM86.7282.0694.7891.28RANSAC49.7052.3583.6553.85LGC81.2265.1292.4089.05Ours90.33 (+3.61)92.67 (+4.77)96.75 (+1.39)94.26 (+1.17)

Table 7.

The average rotation/translation angle errors (

∘

) on four scenes from the NMNET dataset. Smaller is better. The best results are in bold.

Table 7.

The average rotation/translation angle errors (

∘

) on four scenes from the NMNET dataset. Smaller is better. The best results are in bold.

Lib-WideMain-WideMao-WideScience-WideLMR4.87/7.7413.09/18.810.12/0.900.20/0.60LPM3.82/4.408.30/14.220.20/1.330.15/0.52MTOP1.11/2.8511.93/21.030.16/1.240.18/0.61RFM16.17/3.3910.84/24.100.14/1.210.16/0.52RANSAC7.73/9.3218.15/14.850.39/2.320.38/1.54LGC2.67/4.412.40/14.120.18/1.350.24/0.72Ours0.45/1.331.22/13.170.10/0.840.11/0.36

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

Share and Cite

MDPI and ACS Style

Liu, J.; Sun, K.; Jiang, S.; Li, K.; Tao, W.

MSSF: A Novel Mutual Structure Shift Feature for Removing Incorrect Keypoint Correspondences between Images. Remote Sens. 2023, 15, 926.

https://doi.org/10.3390/rs15040926

AMA Style

Liu J, Sun K, Jiang S, Li K, Tao W.

MSSF: A Novel Mutual Structure Shift Feature for Removing Incorrect Keypoint Correspondences between Images. Remote Sensing. 2023; 15(4):926.

https://doi.org/10.3390/rs15040926

Chicago/Turabian Style

Liu, Juan, Kun Sun, San Jiang, Kunqian Li, and Wenbing Tao.

2023. "MSSF: A Novel Mutual Structure Shift Feature for Removing Incorrect Keypoint Correspondences between Images" Remote Sensing 15, no. 4: 926.

https://doi.org/10.3390/rs15040926

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Article Access Statistics

For more information on the journal statistics, click here.

Multiple requests from the same IP address are counted as one view.

Zoom

Orient

As Lines

As Sticks

As Cartoon

As Surface

Previous Scene

Next Scene

Cite

Export citation file:

BibTeX |

EndNote |

RIS

MDPI and ACS Style

Liu, J.; Sun, K.; Jiang, S.; Li, K.; Tao, W.

MSSF: A Novel Mutual Structure Shift Feature for Removing Incorrect Keypoint Correspondences between Images. Remote Sens. 2023, 15, 926.

https://doi.org/10.3390/rs15040926

AMA Style

Liu J, Sun K, Jiang S, Li K, Tao W.

MSSF: A Novel Mutual Structure Shift Feature for Removing Incorrect Keypoint Correspondences between Images. Remote Sensing. 2023; 15(4):926.

https://doi.org/10.3390/rs15040926

Chicago/Turabian Style

Liu, Juan, Kun Sun, San Jiang, Kunqian Li, and Wenbing Tao.

2023. "MSSF: A Novel Mutual Structure Shift Feature for Removing Incorrect Keypoint Correspondences between Images" Remote Sensing 15, no. 4: 926.

https://doi.org/10.3390/rs15040926

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

clear

Remote Sens.,

EISSN 2072-4292,

Published by MDPI

RSS

Content Alert

Further Information

Article Processing Charges

Pay an Invoice

Open Access Policy

Contact MDPI

Jobs at MDPI

Guidelines

For Authors

For Reviewers

For Editors

For Librarians

For Publishers

For Societies

For Conference Organizers

MDPI Initiatives

Sciforum

MDPI Books

Preprints.org

Scilit

SciProfiles

Encyclopedia

JAMS

Proceedings Series

Follow MDPI

Facebook

Twitter

Subscribe to receive issue release notifications and newsletters from MDPI journals

Acoustics

Acta Microbiologica Hellenica

Actuators

Administrative Sciences

Adolescents

Advances in Respiratory Medicine

Aerobiology

Aerospace

Agriculture

AgriEngineering

Agrochemicals

Agronomy

Air

Algorithms

Allergies

Alloys

Analytica

Analytics

Anatomia

Anesthesia Research

Animals

Antibiotics

Antibodies

Antioxidants

Applied Biosciences

Applied Mechanics

Applied Microbiology

Applied Nano

Applied Sciences

Applied System Innovation

AppliedChem

AppliedMath

Aquaculture Journal

Architecture

Arthropoda

Arts

Astronomy

Atmosphere

Atoms

Audiology Research

Automation

Axioms

Bacteria

Batteries

Behavioral Sciences

Beverages

Big Data and Cognitive Computing

BioChem

Bioengineering

Biologics

Biology

Biology and Life Sciences Forum

Biomass

Biomechanics

BioMed

Biomedicines

BioMedInformatics

Biomimetics

Biomolecules

Biophysica

Biosensors

BioTech

Birds

Blockchains

Brain Sciences

Buildings

Businesses

Cancers

Cardiogenetics

Catalysts

Cells

Ceramics

Challenges

ChemEngineering

Chemistry

Chemistry Proceedings

Chemosensors

Children

Chips

CivilEng

Clean Technologies

Climate

Clinical and Translational Neuroscience

Clinics and Practice

Clocks & Sleep

Coasts

Coatings

Colloids and Interfaces

Colorants

Commodities

Complications

Compounds

Computation

Computer Sciences & Mathematics Forum

Computers

Condensed Matter

Conservation

Construction Materials

Corrosion and Materials Degradation

Cosmetics

COVID

Crops

Cryptography

Crystals

Current Issues in Molecular Biology

Current Oncology

Dairy

Data

Dentistry Journal

Dermato

Dermatopathology

Designs

Diabetology

Diagnostics

Dietetics

Digital

Disabilities

Diseases

Diversity

DNA

Drones

Drugs and Drug Candidates

Dynamics

Earth

Ecologies

Econometrics

Economies

Education Sciences

Electricity

Electrochem

Electronic Materials

Electronics

Emergency Care and Medicine

Encyclopedia

Endocrines

Energies

Eng

Engineering Proceedings

Entropy

Environmental Sciences Proceedings

Environments

Epidemiologia

Epigenomes

European Burn Journal

European Journal of Investigation in Health, Psychology and Education

Fermentation

Fibers

FinTech

Fire

Fishes

Fluids

Foods

Forecasting

Forensic Sciences

Forests

Fossil Studies

Foundations

Fractal and Fractional

Fuels

Future

Future Internet

Future Pharmacology

Future Transportation

Galaxies

Games

Gases

Gastroenterology Insights

Gastrointestinal Disorders

Gastronomy

Gels

Genealogy

Genes

Geographies

GeoHazards

Geomatics

Geosciences

Geotechnics

Geriatrics

Gout, Urate, and Crystal Deposition Disease

Grasses

Hardware

Healthcare

Hearts

Hemato

Hematology Reports

Heritage

Histories

Horticulturae

Hospitals

Humanities

Humans

Hydrobiology

Hydrogen

Hydrology

Hygiene

Immuno

Infectious Disease Reports

Informatics

Information

Infrastructures

Inorganics

Insects

Instruments

International Journal of Environmental Research and Public Health

International Journal of Financial Studies

International Journal of Molecular Sciences

International Journal of Neonatal Screening

International Journal of Plant Biology

International Journal of Translational Medicine

International Journal of Turbomachinery, Propulsion and Power

International Medical Education

Inventions

IoT

ISPRS International Journal of Geo-Information

Journal of Ageing and Longevity

Journal of Cardiovascular Development and Disease

Journal of Clinical & Translational Ophthalmology

Journal of Clinical Medicine

Journal of Composites Science

Journal of Cybersecurity and Privacy

Journal of Developmental Biology

Journal of Experimental and Theoretical Analyses

Journal of Functional Biomaterials

Journal of Functional Morphology and Kinesiology

Journal of Fungi

Journal of Imaging

Journal of Intelligence

Journal of Low Power Electronics and Applications

Journal of Manufacturing and Materials Processing

Journal of Marine Science and Engineering

Journal of Market Access & Health Policy

Journal of Molecular Pathology

Journal of Nanotheranostics

Journal of Nuclear Engineering

Journal of Otorhinolaryngology, Hearing and Balance Medicine

Journal of Personalized Medicine

Journal of Pharmaceutical and BioTech Industry

Journal of Respiration

Journal of Risk and Financial Management

Journal of Sensor and Actuator Networks

Journal of Theoretical and Applied Electronic Commerce Research

Journal of Vascular Diseases

Journal of Xenobiotics

Journal of Zoological and Botanical Gardens

Journalism and Media

Kidney and Dialysis

Kinases and Phosphatases

Knowledge

Laboratories

Land

Languages

Laws

Life

Limnological Review

Lipidology

Liquids

Literature

Livers

Logics

Logistics

Lubricants

Lymphatics

Machine Learning and Knowledge Extraction

Machines

Macromol

Magnetism

Magnetochemistry

Marine Drugs

Materials

Materials Proceedings

Mathematical and Computational Applications

Mathematics

Medical Sciences

Medical Sciences Forum

Medicina

Medicines

Membranes

Merits

Metabolites

Metals

Meteorology

Methane

Methods and Protocols

Metrology

Micro

Microbiology Research

Micromachines

Microorganisms

Microplastics

Minerals

Mining

Modelling

Molbank

Molecules

Multimodal Technologies and Interaction

Muscles

Nanoenergy Advances

Nanomanufacturing

Nanomaterials

NDT

Network

Neuroglia

Neurology International

NeuroSci

Nitrogen

Non-Coding RNA

Nursing Reports

Nutraceuticals

Nutrients

Obesities

Oceans

Onco

Optics

Oral

Organics

Organoids

Osteology

Oxygen

Parasitologia

Particles

Pathogens

Pathophysiology

Pediatric Reports

Pharmaceuticals

Pharmaceutics

Pharmacoepidemiology

Pharmacy

Philosophies

Photochem

Photonics

Phycology

Physchem

Physical Sciences Forum

Physics

Physiologia

Plants

Plasma

Platforms

Pollutants

Polymers

Polysaccharides

Poultry

Powders

Proceedings

Processes

Prosthesis

Proteomes

Psych

Psychiatry International

Psychoactives

Publications

Quantum Beam Science

Quantum Reports

Quaternary

Radiation

Reactions

Real Estate

Receptors

Recycling

Religions

Remote Sensing

Reports

Reproductive Medicine

Resources

Rheumato

Risks

Robotics

Ruminants

Safety

Sci

Scientia Pharmaceutica

Sclerosis

Seeds

Sensors

Separations

Sexes

Signals

Sinusitis

Smart Cities

Social Sciences

Société Internationale d’Urologie Journal

Societies

Software

Soil Systems

Solar

Solids

Spectroscopy Journal

Sports

Standards

Stats

Stresses

Surfaces

Surgeries

Surgical Techniques Development

Sustainability

Sustainable Chemistry

Symmetry

SynBio

Systems

Targets

Taxonomy

Technologies

Telecom

Textiles

Thalassemia Reports

Thermo

Tomography

Tourism and Hospitality

Toxics

Toxins

Transplantology

Trauma Care

Trends in Higher Education

Tropical Medicine and Infectious Disease

Universe

Urban Science

Uro

Vaccines

Vehicles

Venereology

Veterinary Sciences

Vibration

Virtual Worlds

Viruses

Vision

Waste

Water

Wind

Women

World

World Electric Vehicle Journal

Youth

Zoonotic Diseases

Disclaimer

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely

those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or

the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas,

methods, instructions or products referred to in the content.

Terms and Conditions

We use cookies on our website to ensure you get the best experience.

GitHub - fkxxyz/ssfconv: Sogou input method skin file (.ssf file) converter, supports conversion to fcitx or fcitx5 format.

Toggle navigation

Product

Actions

Automate any workflow

Packages

Host and manage packages

Security

Find and fix vulnerabilities

Codespaces

Instant dev environments

Copilot

Write better code with AI

Code review

Manage code changes

Issues

Plan and track work

Discussions

Collaborate outside of code

Explore

All features

Documentation

GitHub Skills

Blog

Solutions

For

Enterprise

Teams

Startups

Education

By Solution

CI/CD & Automation

DevOps

DevSecOps

Resources

Learning Pathways

White papers, Ebooks, Webinars

Customer Stories

Partners

Open Source

GitHub Sponsors

Fund open source developers

The ReadME Project

GitHub community articles

Repositories

Topics

Trending

Collections

Pricing

Search or jump to...

Search code, repositories, users, issues, pull requests...

Clear

Search syntax tips

Provide feedback

We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Cancel

Submit feedback

Saved searches

Use saved searches to filter your results more quickly

Name

Query

To see all available qualifiers, see our documentation.

Cancel

Create saved search

You signed in with another tab or window. Reload to refresh your session.

You signed out in another tab or window. Reload to refresh your session.

You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

fkxxyz

ssfconv

Public

Notifications

Fork

Star

172

Sogou input method skin file (.ssf file) converter, supports conversion to fcitx or fcitx5 format.

License

GPL-3.0 license

172

stars

forks

Branches

MSSF-GCN: Multi-scale Structural and Semantic Information Fusion Graph Convolutional Network for Controversy Detection | SpringerLink

Find a journal

Publish with us

Track your research

Cart

International Conference on Web Information Systems EngineeringWISE 2021: Web Information Systems Engineering – WISE 2021

394–402Cite as

Home

Web Information Systems Engineering – WISE 2021

Conference paper

MSSF-GCN: Multi-scale Structural and Semantic Information Fusion Graph Convolutional Network for Controversy Detection

Haiyang Wang12, Xin Song12, Bin Zhou12, Ye Wang12, Liqun Gao12 & …Yan Jia12 Show authors

Conference paper

First Online: 01 January 2022

1364 Accesses

Part of the Lecture Notes in Computer Science book series (LNISA,volume 13080)

AbstractDetecting controversial posts on the web and social media play an important role in judging the authenticity of web information, measuring the influence of news and alleviating the polarized views. The controversy detection task has attracted widespread attention from researchers in the fields of computer science and social humanities sciences. However, previous works do not achieve: 1) preserve the reply-structure relationship with sentiment information; 2) integrate multi-scale structure and semantic information and provide interpretable results; 3) learn effectively topics and comments information related to the target post. To overcome the first limitation, we construct a Topic-Post-Comment-Sentiment Graph (TPCS Graph) for preserving the reply-structure and incorporate the sentiment information. For the second and third limitation, we propose Multi-scale Structural and Semantic Information Fusion Graph Convolutional Network (MSSF-GCN) for post-level controversy detection. Moreover, we build a multilingual dataset for controversy detection. We conduct comprehensive experiments on two real-world datasets and the results show that the proposed method exhibits comparable or even superior performance.KeywordsControversy detectionInformation fusionGraph convolutional networks

This is a preview of subscription content, log in via an institution.

Buying options

Chapter

EUR 29.95

Price includes VAT (Philippines)

Available as PDF

Read on any device

Instant download

Own it forever

Buy Chapter

eBook

EUR 85.59

Price includes VAT (Philippines)

Available as EPUB and PDF

Read on any device

Instant download

Own it forever

Buy eBook

Softcover Book

EUR 99.99

Price excludes VAT (Philippines)

Compact, lightweight edition

Dispatched in 3 to 5 business days

Free shipping worldwide - see info

Buy Softcover Book

Tax calculation will be finalised at checkout

Purchases are for personal use onlyLearn about institutional subscriptions

Notes1.https://huggingface.co/transformers/task_summary.html.2.http://mcg.ict.ac.cn/controversy-detection-dataset.html. ReferencesConneau, A., Lample, G.: Cross-lingual language model pretraining. In: Advances in Neural Information Processing Systems, vol. 32, pp. 7057–7067 (2019)

Google Scholar

Devlin, J., Chang, M.W., Lee, K., Toutanova, K.N.: BERT: pre-training of deep bidirectional transformers for language understanding. In: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pp. 4171–4186 (2018)

Google Scholar

Donnat, C., Zitnik, M., Hallac, D., Leskovec, J.: Learning structural node embeddings via diffusion wavelets. In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 1320–1329 (2018)

Google Scholar

Dori-Hacohen, S.: Controversy analysis and detection (2017)

Google Scholar

Garimella, K., Morales, G.D.F., Gionis, A., Mathioudakis, M.: Quantifying controversy on social media. ACM Trans. Soc. Comput. 1(1), 3 (2018)Article

Google Scholar

Kingma, D.P., Ba, J.L.: Adam: a method for stochastic optimization. In: ICLR 2015 : International Conference on Learning Representations 2015 (2015)

Google Scholar

Kipf, T.N., Welling, M.: Semi-supervised classification with graph convolutional networks. In: ICLR (Poster) (2016)

Google Scholar

Popescu, A.M., Pennacchiotti, M.: Detecting controversial events from twitter. In: Proceedings of the 19th ACM International Conference on Information and Knowledge Management, pp. 1873–1876 (2010)

Google Scholar

Rethmeier, N., Hübner, M., Hennig, L.: Learning comment controversy prediction in web discussions using incidentally supervised multi-task CNNs. In: Proceedings of the 9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, pp. 316–321 (2018)

Google Scholar

Vamvas, J., Sennrich, R.: X-stance: a multilingual multi-target dataset for stance detection. SwissText/KONVENS (2020)

Google Scholar

Zhong, L., Cao, J., Sheng, Q., Guo, J., Wang, Z.: Integrating semantic and structural information with graph convolutional network for controversy detection. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp. 515–526 (2020)

Google Scholar

Download references AcknowledgementsThis work was supported by the National Key Research and Development Program of China NO. 2018YFC0831703. Author informationAuthors and AffiliationsNational University of Defense Technology, Changsha, ChinaHaiyang Wang, Xin Song, Bin Zhou, Ye Wang, Liqun Gao & Yan JiaAuthorsHaiyang WangView author publicationsYou can also search for this author in

PubMed Google ScholarXin SongView author publicationsYou can also search for this author in

PubMed Google ScholarBin ZhouView author publicationsYou can also search for this author in

PubMed Google ScholarYe WangView author publicationsYou can also search for this author in

PubMed Google ScholarLiqun GaoView author publicationsYou can also search for this author in

PubMed Google ScholarYan JiaView author publicationsYou can also search for this author in

PubMed Google ScholarCorresponding authorCorrespondence to

Bin Zhou . Editor informationEditors and AffiliationsSchool of Computer Science and Engineering, University of New South Wales, Sydney, AustraliaWenjie Zhang Peking University, Beijing, ChinaLei Zou Zayed University, Dubai, United Arab EmiratesZakaria Maamar Swinburne University of Technology, Melbourne, VIC, AustraliaLu Chen Rights and permissionsReprints and permissions Copyright information© 2021 Springer Nature Switzerland AG About this paperCite this paperWang, H., Song, X., Zhou, B., Wang, Y., Gao, L., Jia, Y. (2021). MSSF-GCN: Multi-scale Structural and Semantic Information Fusion Graph Convolutional Network for Controversy Detection.

In: Zhang, W., Zou, L., Maamar, Z., Chen, L. (eds) Web Information Systems Engineering – WISE 2021. WISE 2021. Lecture Notes in Computer Science(), vol 13080. Springer, Cham. https://doi.org/10.1007/978-3-030-90888-1_30Download citation.RIS.ENW.BIBDOI: https://doi.org/10.1007/978-3-030-90888-1_30Published: 01 January 2022

Publisher Name: Springer, Cham

Print ISBN: 978-3-030-90887-4

Online ISBN: 978-3-030-90888-1eBook Packages: Computer ScienceComputer Science (R0)Share this paperAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard

Provided by the Springer Nature SharedIt content-sharing initiative

Publish with usPolicies and ethics

Access via your institution

Buying options

Chapter

EUR 29.95

Price includes VAT (Philippines)

Available as PDF

Read on any device

Instant download

Own it forever

Buy Chapter

eBook

EUR 85.59

Price includes VAT (Philippines)

Available as EPUB and PDF

Read on any device

Instant download

Own it forever

Buy eBook

Softcover Book

EUR 99.99

Price excludes VAT (Philippines)

Compact, lightweight edition

Dispatched in 3 to 5 business days

Free shipping worldwide - see info

Buy Softcover Book

Tax calculation will be finalised at checkout

Purchases are for personal use onlyLearn about institutional subscriptions

Search by keyword or author

Navigation

Find a journal

Publish with us

Track your research

Discover content

Journals A-Z

Books A-Z

Publish with us

Publish your research

Open access publishing

Products and services

Our products

Librarians

Societies

Partners and advertisers

Our imprints

Springer

Nature Portfolio

BMC

Palgrave Macmillan

Apress

Your privacy choices/Manage cookies

Your US state privacy rights

Accessibility statement

Terms and conditions

Help and support

49.157.13.121

Not affiliated

Exploring flavour-producing core microbiota in multispecies solid-state fermentation of traditional Chinese vinegar | Scientific Reports

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain

the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in

Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles

and JavaScript.

View all journals

Explore content

About the journal

Publish with us

RSS feed

nature

scientific reports

articles

article

Exploring flavour-producing core microbiota in multispecies solid-state fermentation of traditional Chinese vinegar

Download PDF

Article

Open access

Published: 31 May 2016

Exploring flavour-producing core microbiota in multispecies solid-state fermentation of traditional Chinese vinegar

Zong-Min Wang1 na1, Zhen-Ming Lu1,2 na1, Jin-Song Shi1,3 na1 & …Zheng-Hong Xu1,2,3 na1 Show authors

Scientific Reports

volume 6, Article number: 26818 (2016)

Cite this article

7071 Accesses

123 Citations

9 Altmetric

Metrics details

Subjects

Data integrationFood microbiologyMicrobial ecology

AbstractMultispecies solid-state fermentation (MSSF), a natural fermentation process driven by reproducible microbiota, is an important technique to produce traditional fermented foods. Flavours, skeleton of fermented foods, was mostly produced by microbiota in food ecosystem. However, the association between microbiota and flavours and flavour-producing core microbiota are still poorly understood. Here, acetic acid fermentation (AAF) of Zhenjiang aromatic vinegar was taken as a typical case of MSSF. The structural and functional dynamics of microbiota during AAF process was determined by metagenomics and favour analyses. The dominant bacteria and fungi were identified as Acetobacter, Lactobacillus, Aspergillus and Alternaria, respectively. Total 88 flavours including 2 sugars, 9 organic acids, 18 amino acids and 59 volatile flavours were detected during AAF process. O2PLS-based correlation analysis between microbiota succession and flavours dynamics showed bacteria made more contribution to flavour formation than fungi. Seven genera including Acetobacter, Lactobacillus, Enhydrobacter, Lactococcus, Gluconacetobacer, Bacillus and Staphylococcus were determined as functional core microbiota for production of flavours in Zhenjiang aromatic vinegar, based on their dominance and functionality in microbial community. This study provides a perspective for bridging the gap between the phenotype and genotype of ecological system and advances our understanding of MSSF mechanisms in Zhenjiang aromatic vinegar.

Similar content being viewed by others

Dissecting industrial fermentations of fine flavour cocoa through metagenomic analysis

Article

Open access

21 April 2021

Miguel Fernández-Niño, María Juliana Rodríguez-Cubillos, … Andrés Fernando González Barrios

Integrated meta-omics approaches reveal Saccharopolyspora as the core functional genus in huangjiu fermentations

Article

Open access

19 September 2023

Shuangping Liu, Zhi-Feng Zhang, … Jian Mao

Evaluation of microbial communities of Chinese Feng-flavor Daqu with effects of environmental factors using traceability analysis

Article

Open access

11 May 2023

Yongli Zhang, Chen Xu, … Yaodong Chen

IntroductionMultispecies solid-state fermentation (MSSF), is defined as a fermentation process in which multiple microorganisms grow on solid-state materials without present of free liquid. It might be one of the oldest and most economical ways of producing and preserving foods. It has been proved MSSF may improve the nutritional value, taste, smell and healthy function of raw materials1,2. This traditional fermentation method is maintained through a spontaneous mixed-culture refreshment process without sterilisation. Enhanced by repeated practices for years, specific microbiota have been well characterised and their potential in food industry has been exploited intentionally3,4,5. It can be concluded the success of MSSF could rely on the reproducible formation of well-balanced microbiota, which determines the safety, smell, taste, texture and aroma of fermented foods.With the development of ecological techniques there are increasing studies to investigate food fermentation, focusing on the patterns/dynamics of the multi-species microbiota4,5,6,7 and the functionality of the microbial community6,8,9,10. These studies provide crucial information to help understand the role of microbiota and the function of the community in fermented foods. However, due to the complexity of MSSF and the lack of data mining strategy, the correlation between microbiota and flavours is still not clear11. Moreover, how to pick indicative functional core microbes from high species community, taking into account both dominance and functionality, is still challenging. Along with the advance of next generation sequencing, the principal research burdens are transforming from traditional wet-lab experiments to dealing with huge and informative data12. Bidirectional orthogonal partial least squares (O2PLS) method is an efficient statistic approach to integrate data collected from different analytical platform and dig into the potential associations between two disparate datasets13. This approach has been applied to investigate the metabolomic and proteomic correlation from mice samples14, the microbes and metabolic phenotype correlation in human gut15 and integrate transcript and metabolite data in plant biology16. However, there were scarce studies to inquire into associations between different omics platforms in fermented foods.Zhenjiang aromatic vinegar, a well-known traditional fermented vinegar, is produced by three major steps including alcohol fermentation, acetic acid fermentation (AAF) and aging. Hereinto, AAF is a typical MSSF process with alcohol mash, wheat bran and chaff as raw materials and fermented cereals from the last batch of AAF (termed Pei in Chinese, inoculum size, 8%, w/w) as starter17 (Fig. S1b). The succession of microbiota in the Pei during AAF process results in a dynamic flavours composition, which directly affects the taste and aroma of vinegars. Variation of flavours in fermented vinegar has been extensively studied by nuclear magnetic resonance spectroscopy, raman spectroscopy and mass spectrometry18,19,20,21,22,23. The microbial ecology during AAF process has also been investigated by culture-based and culture-independent approaches4,5,6,7. However, the correlation between microbiota and flavours and flavour-producing core microbiota remain to be determined in fermented vinegars.To address this challenge, the assembly and dynamics of microbiota in vinegar Pei during AAF process were characterised by MiSeq sequencing. The changes of flavours composition during AAF were detected by chromatography and analysed by multivariable statistics. Based on these information, the relationship between microbiota assembly and flavours datasets was investigated by O2PLS. Finally, a functional core microbiota was selected by comparison of the comprehensive importance of microbiota correlated with flavours during AAF process.ResultsPhylogenetic landscapes and dynamics of microbiota during AAF processPCR-based amplicon sequencing was applied to characterise the microbiota assembly and dynamics in vinegar Pei during AAF process. Across all samples, total 253 and 657 operational taxonomic units (OTUs) were detected for bacteria and fungi respectively with 97% similarity. The average of Good’s coverage was over 0.99 for all samples (Dataset S1), indicating the identified sequences represented majority of microbiota in vinegar Pei. Bacterial assembly were dominated by Firmicutes and Proteobacteria, while the fungi predominantly consisted of the phyla Ascomycota, Fungi_unclassified and Basidiomycota (Fig. S2). A total of 151 bacterial genera and 202 fungal genera were identified in vinegar Pei during AAF process. As for bacteria, Lactobacillus was predominant in the early stage of AAF (days 0–9), while Acetobacter, Lactococcus, Gluconacetobacter, Enterococcus and Bacillus were prevailing in the later stage of AAF (days 10–18). Therein Acetobacter could mainly originate from the starter cultures (#v_sp in Fig. 1a) and Lactococcus could mainly originate from alcohol mash (#v_am). Gluconacetobacter, Enterococcus and Bacillus might originate from the raw materials (#v_mp), which were increasing with the proceeding of AAF (Fig. 1a). As for fungi, Aspergillus was existed in the whole AAF process, which increased in the early 13 days of AAF and then maintained fluctuation in small range (0.4–0.5). Alternaria was dominated in early stage of AAF (days 1–6) and then decreased gradually to the end of AAF. Fungi_unclassified accounted for more than 60% in sample #day_0 but declined rapidly once AAF started, which might originate from the alcohol mash (#v_am) and raw materials (#v_mp) (Fig. 1a). A total of 21 yeast genera were identified in vinegar Pei, including Cryptococcus, Debaryomyces, Candida, Saccharomyces and so on (Fig. S3). However, these genera only accounted for 1.6% in fungal community. The biomass of bacteria was increasing in the first 7 days and then decreased till the end of AAF while the biomass of fungi increased in the first 4 days and then decreased till the end of AAF (Fig. 1b). Moreover, the biomass ratio of bacteria and fungi was in the range of 165 to 13,300, which suggested the bacteria played key role in the solid-state AAF.Figure 1Distribution of microbiota in vinegar Pei and biomass of the bacteria and fungi in different samples during AAF process.(a) Average distribution of bacterial and fungal genera in vinegar Pei during AAF process. (b) Average biomass analysis of bacteria and fungi in vinegar Pei during AAF process.Full size imageComparison of the microbiota structure in vinegar Pei between different AAF stagesThough the dominant genera such as Acetobacter, Lactobacillus and Aspergillus were widely distributed across the vinegar Pei, their abundance within each sample is variable. Principal component analysis (PCA) was applied to compare the microbiota of vinegar Pei in different stages of AAF. It was shown that both bacterial and fungal community structure of the samples on day 0 of AAF exhibited little similarity to other samples except for raw material samples (#v_am and #v_mp) (Fig. 2a,b). The samples in early stage were clustered separately from the samples in late stage of AAF based on the assembly and variation of microbiota, which indicated AAF process could be divided into three stages: I, day 0 (red circle in Fig. 2); II, days 1–9 (green box in Fig. 2); and III, days10–18 (blue triangle in Fig. 2). Furthermore, AMOVA showed that the degree of variation (Fs) among all stages was larger than within stages and p-value between any two stages of AAF (I vs. II, I vs. III and II vs. III) was less than 0.001, suggesting the comparison of the divided three stages during AAF process was statistically significant (Fig. 2a,b). As for bacterial community, metastats analysis revealed a total of 38 OTUs, 21 OTUs and 52 OTUs in stage I, II and III of AAF were significantly different from other two stages (p < 0.05) respectively. Therein, Pseudomonas, Methylobacterium, Lactobacillus, Sphingomonas, Rhizobium, Staphylococcus, Xanthomonas and Acetobacter were significant different genera in three stages. As for fungal community, it was shown that total 40 OTUs, 29 OTUs and 25 OTUs in stage I, II and III of AAF were significantly different from other two stages (p < 0.05) respectively, where Aspergillus, Verticillium, Rhizomucor, Fungi_unclassified, Pleosporales_unclassified and Eurotiales_unclassified were significant different genera in three groups. Details of the bacterial and fungal taxonomy classification of the significant OTUs are listed in Tables S1, S2 and Dataset S1. In addition, the acidic stress and alcohol stress were two best predictors of bacterial and fungal community composition, with the principal coordinate one (PC1) being significantly associated with the gradient of titratable acidity and alcohol during AAF process (Fig. 2c, Bacteria: titratable acidity (rho, 0.906), alcohol (rho, −0.901); Fungi: titratable acidity (rho, 0.509), alcohol (rho, −0.545)), but the gradient of temperature was nearly not correlated with the bacterial and fungal community composition (Fig. S4, Bacteria: rho, −0.0974; Fungi: rho, −0.0859).Figure 2Comparison of the structure of microbiota in different samples and correlation between microbiota and environmental factors during AAF process.(a) PCA and AMOVA results of bacterial community in vinegar Pei at different stages of AAF based on hellinger distance with 97% similarity. (b) PCA and AMOVA results of fungal community in vinegar Pei at different stages of AAF based on hellinger distance with 97% similarity. (c) Correlation between the first principal component (PC1, bacteria and fungi) and titratable acidity and alcohol level respectively.Full size imageMultivariate analysis of flavours during AAF processA total of 88 flavours were detected during AAF process, including 2 sugars, 9 organic acids (OAs), 18 amino acids (AAs) and 59 volatile flavours (VFs). The volatile flavours could be divided into seven categories including 9 alcohols (No. 1–9), 8 acids (No. 10–17), 25 esters (No. 18–42), 4 ketones (No. 43–46), 7 aldehydes (No. 47–53), 3 heterocycles (No. 54–56) and 3 others (No. 57–59) (Fig. S5). PCA analysis showed that the first two components R2X(cum) explained 63.2% of the variables and the cross-validated Q2-value for each component were more than the cross validation threshold for that component (Limit), indicating significant components for this analysis (Dataset S2). The projected coordinate of metabolites in PC1 appeared to capture the evolutionary tendency of flavours during AAF process and dynamics of flavours were clearly distinct in different stages of AAF (Fig. 3b). Hierarchical cluster analysis (HCA) revealed the AAF process could be divided into 3 groups based on flavours: group1, day 0; group 2, days 1–7; group 3, days 8–18 (labelled red, green and blue in Fig. 3a respectively). A biplot integrating scores and loadings demonstrated there were 13 flavours including fructose, glucose and 11 VFs highly correlated with group 1 (red circle in Fig. 3b); 15 VFs highly correlated with group 2 (green box in Fig. 3b); and 60 flavours including 9 OAs, 18 AAs and 33 VFs highly correlated with group 3 (blue triangle in Fig. 3b). More detailed information is provided in Table S3. These results suggested most of the flavours (OAs, AAs and half of VFs) were produced in the late stages of AAF (days 8–18).Figure 3PCA and HCA analysis of flavours in vinegar Pei during AAF process.(a) The dendrogram of AAF process was obtained by hierarchical cluster analysis based on PCA modeling. (b) The biplot superimposed the scores and loadings of PCA analysis based on correlation scaling method. R2VX represents the fraction of X variation modeled in the component. p(corr), t(corr) is a combined vector, p(corr) represents loading p scaled as correlation coefficient between X and t; t(corr) represents score t scaled as correlation coefficient resulting in all points falling inside the circle with radius 1.Full size imageAssociation between microbiota and flavours during AAF processO2PLS method was used to analyse the association between microbiota and flavours during AAF process. It was shown R2 and Q2 of the model was 0.879 and 0.528 respectively (Dataset S2, Fig. S6a), suggesting O2PLS method was well fitted for analysis and prediction. The first two predictive components were significant by cross validation, accounting for 90% of R2(cum) and 100% of Q2(cum) in this model (Fig. S6b). The VIP(pred) vector (VIP value for the predictive components) of analysed microbiota varied in 0.15–1.63, in which total 85 microbial genera (VIP(pred) > 1.0) including 66 bacterial genera (VIP(pred) ≈ 1.03–1.63) and 19 fungal genera (VIP(pred) ≈ 1.01–1.46) had important effects on the flavours (Fig. 4a, Dataset S2), suggesting bacteria were more important for vinegar production than fungi. Acetobacter, Lactobacillus, Gluconacetobacter and Lactococcus were the biggest contributors to the production of flavours during AAF process. Based on correlation coefficient between microbiota and flavours, a total of 94 genera including 61 bacteria (green circles in left side of Fig. 4b) and 33 fungi (yellow circles in left side of Fig. 4b) were moderately and highly correlated (|ρ| > 0.7) with all three flavour sets, in which total 47 genera (36 bacteria and 11 fungi) were correlated with OAs (light red circles in right side of Fig. 4b); 59 genera (48 bacteria and 11 fungi) were correlated with AAs (light green circles in right side of Fig. 4b); and 92 genera (61 bacteria and 31 fungi) were correlated with VFs (light blue labels in right side of Fig. 4b). Acetobacter and Lactobacillus possessed the largest number of correlated flavours (56 and 53 respectively), while Aspergillus and Fungi_unclassified were correlated with 39 and 34 of flavours respectively (|ρ| > 0.7) (Table S8). Most of fungal genera (75.7%) had correlated with few flavours (≤5), in which 14 genera had poor correlated with only one flavour. Details of the relationships between the microbiota and flavours are listed in Table S4 and Table S8.Figure 4Correlation analyses between microbiota and flavours by O2PLS modeling during AAF process.(a) VIP(pred) (variable importance for predictive components) plot of the important microbiota (VIP(pred) > 1.0). (b) The correlated network between microbial genera and flavours during AAF process. The left-side circles represent the bacterial (green) and fungal (yellow) genera correlated with flavours (|ρ| > 0.7). The right-side circles represent the flavours (sugars, light purple circle; organic acids, light red circle; amino acids, light green circle; volatile flavours, light blue labels (hexagons: alcohols; octagons: acids; circles: esters; diamonds: ketones; vees: aldehydes; rects: heterocycles; triangles: others)) correlated with microbiota (|ρ| > 0.7). The red long dashed lines linking the circles represent positive correlation while the blue long dashed lines represent the negative correlation between microbiota and flavours.Full size imageFor OAs, bacteria played more important role than fungi, in which Lactobacillus, Enhydrobacter and Gluconacetobacter were important genera for the production of OAs during AAF process. Acetic acid (AA) and lactic acid (LA) were main acids in cereal vinegar. Total 25 genera were correlated with AA (|ρ| > 0.7) (Fig. 4b, Table S5), in which Acetobacter, Enhydrobacter and Lactobacillus had excellent correlation with AA (|ρ| > 0.9), indicating the three genera were mainly responsible for the change of AA during AAF process. LA was positively correlated with Phaeoseptoria and Fusarium; and negatively correlated with 14 genera during AAF process. Therein Staphylococcus and Weissella were two most important genera for change of LA during AAF process. Detailed information of genera correlated with each organic acid is summarised in Table S5.For AAs, Acetobacter, Aspergillus, Lactobacillus, Enhydrobacter, Roseomonas, Sphingobacterium, Staphylococcus, Stenotrophomonas and Fungi_unclassified were crucial to dynamics of AAs during AAF process (Fig. 4b). Glutamic acid (Glu), alanine (Ala), valine (Val) and leucine (Leu) are four abundant flavours for the taste of vinegar. Glu and Leu, providing umami and bitter taste of vinegar, were correlated with 14 and 22 genera (|ρ| > 0.7) respectively, in which Staphylococcus, Acetobacter, Sphingobacterium and Aspergillus were highly correlated with the changes of Glu and Leu during AAF process (|ρ| > 0.8). Ala, providing sweet taste of vinegar, was correlated with 16 genera (|ρ| > 0.7), in which Acetobacter, Aspergillus, Staphylococcus and Lactobacillus were most important (|ρ| > 0.8) for the change of Ala during AAF process. Val, providing sweet and bitter taste of vinegar, was correlated with 14 genera (|ρ| > 0.7), in which Acetobacter, Aspergillus, Sphingobacterium and Staphylococcus were the major Val producers (|ρ| > 0.8). Moreover, γ-aminobutyric acid (Gaba), a bioactive component in vinegar, has physiological functions to depress the elevation of systolic blood pressure24. Change of Gaba during AAF process was correlated with 7 genera (|ρ| > 0.6), in which Epicoccum and Alternaria were the most important genera. Details of correlated genera with each amino acid are listed in Table S6.Acetobacter, Lactococcus, Lactobacillus and Gluconacetobacer were important to dynamics of VFs during AAF process, which were correlated with more than 30 VFs (|ρ| > 0.7) (Fig. 4b, Table S8). A total of 56 genera were correlated with 9 volatile alcohols (|ρ| > 0.7), in which Acetobacter, Lactobacillus, Enhydrobacter, Lactococcus, Bacillales_unclassified, Gluconacetobacer, Enterococcus, Arthrobacter, Carnobacterium, Verticillium and Nitriliruptor were correlated with more than 7 alcohols (light blue hexagons in Fig. 4b). Total 51 genera were correlated with 8 volatile acids and most of the correlation were positive (|ρ| > 0.7) (light blue octagons in Fig. 4b). There were 81 genera correlated with 25 volatile esters (|ρ| > 0.7), in which most of fungal genera were correlated with few esters (≤4) (light blue circles in Fig. 4b). There were 27 genera correlated with 4 volatile ketones (|ρ| > 0.7) and most of the correlation were positive (light blue diamonds in Fig. 4b). There were 48 genera correlated with 7 volatile aldehydes (|ρ| > 0.7), in which Staphylococcus and Sphingobacterium were correlated with more than 5 aldehydes (light blue vees in Fig. 4b). Total 21 genera were correlated with 3 volatile heterocycles and the correlation are positive except Lactobacillus (light blue rects in Fig. 4b). 2,3,5,6-tetramethyl-pyrazine (No. 56, known as ligustrazine), a functional bioactivator in vinegar, was correlated with 16 genera, in which Gluconacetobacer, Ruminococcaceae_unclassified and Sphingobium were excellently correlated with change of ligustrazine (|ρ| > 0.9). Total 50 genera were correlated with 3 others volatile (|ρ| > 0.7)(light blue triangles in Fig. 4b). In addition, there were 9 volatile flavours exhibited a weak correlation with microbiota (|ρ| < 0.7), suggesting these metabolites might be produced by natural physiochemical process. Details of the microbiota correlated with each volatile flavour are listed in Table S7.Analysis of the functional core microbiota for vinegar fermentationFurther analysis was performed to investigate the relationship of microbiota highly correlated with three flavour sets in vinegar Pei during AAF process (|ρ| > 0.8) (Fig. 5a, Table S9). It was shown there were 23, 37 and 62 genera highly correlated with OAs, AAs and VFs respectively. Total 21 genera including 19 bacterial genera and 2 fungal genera were common to three flavour sets. Fungal genera highly correlated with VFs were more than OAs and AAs, which indicated the fungal community in vinegar Pei were partly contributed to the aroma and fragrance of vinegar. In order to study the functional core microbiota in vinegar Pei, several conditions should be considered: (i) detected stably in AAF process; (ii) the shared microbiota among three flavour sets; (iii) the VIP(pred) value of microbe was greater than 1.55; (iv) the number of flavours highly correlated with microbiota (|ρ| > 0.8) was greater than 25. Based on these, seven genera including Acetobacter (G1), Lactobacillus (G2), Enhydrobacter (G3), Lactococcus (G4), Gluconacetobacer (G6), Bacillus (G7) and Staphylococcus (G10) were selected as functional core microbiota for AAF of Zhenjiang aromatic vinegar (Fig. 5b). These seven genera were highly correlated with dynamics of 69 flavours during AAF process, including 9 OAs, 16 AAs and 44 VFs (|ρ| > 0.8). Therein, Acetobacter, Gluconacetobacer, Lactobacillus and Enhydrobacter were mainly responsible for the change of OAs, while Acetobacter and Staphylococcus were mainly responsible for the change of AAs. These 7 genera were all contributed to the change of VFs, in which Acetobacter, Lactobacillus and Enhydrobacter were mainly responsible for that of volatile alcohols; Gluconacetobacer was mainly responsible for changes of volatile esters and heterocycles; and Staphylococcus was mainly responsible for volatile aldehydes. More detailed information about the functional core microbiota is listed in Table S10. In addition, PICRUSt analysis revealed that the predicted functions of the core microbiota and non-core microbiota were all assigned to seven categories including metabolism, unclassified, genetic information processing, environmental information processing, organismal systems, cellular processes and none (Fig. S7). Therein, metabolism was the main function (39.11%) of the microbial community in vinegar Pei, mainly including amino acid metabolism, carbohydrate metabolism and biosynthesis of other secondary metabolites (Dataset S2). The core microbiota could contribute to 87.87% of metabolism function (Fig. S7). Genetic information processing and environmental information processing were essential to the microbial community in vinegar Pei (occupied 21.49% and 13.33% respectively), which were also mainly carried out (88.60%) by core microbiota (Fig. S7). These suggested the core microbiota could perform the most function of the total microbial community in vinegar production.Figure 5Analysis of the core microbiota for vinegar Pei during AAF process.(a) Venn diagram of relationship of microbiota highly correlated with organic acids, amino acids and volatile flavours (|ρ| > 0.8). (b) The core microbiota accord with the following terms: (i) detected stably in the whole process of AAF; (ii) the shared microbiota for three flavour sets; (iii) the VIP(pred) value of microbiota was greater than 1.55; (iv) the number of flavours highly correlated with microbiota (|ρ| > 0.8) was greater than 25.Full size imageDiscussionMicrobiota inhabiting in vinegar Pei is of great importance for the quality and characteristics of cereal vinegars. Many molecular ecological approaches have been used to characterise the bacterial and fungal community4,5,25,26,27. In this study, 151 bacterial genera and 202 fungal genera in vinegar Pei during AAF process were identified by next generation sequencing, revealing higher diversity and quantitative abundance than previous studies4,5,28,29,30,31,32. The majority of sequences in vinegar Pei were assigned to Acetobacter, Lactobacillus, Aspergillus and Alternaria, which were consisted with the previous studies4,26. Acetobacter was increased during AAF process while Lactobacillus was gradually decreased. This succession tendency might be as a potential indicator to ensure the normal AAF process. Yeast community in vinegar Pei included 21 identified genera in this study, suggesting higher diversity than that in Tianjin duliu mature vinegar26 and traditional balsamic vinegar33. However, the abundance of yeast was only occupied 1.6% in fungal genera and the conjecture was yeast autolysis occurred after alcohol fermentation34,35. During manufacturing, the AAF process is controlled empirically and the complexity of microbiota make it difficult to be used as a rational approach to monitor AAF process. Here, Miseq sequencing provided well depth to cover the complex microbiota in vinegar Pei. Based on structure of microbiota, the AAF process was divided into three distinct stages: I, day 0; II, days 1–9; and III, days 10–18. This division provided a succession profile of microbiota during AAF, which could be used to search for microbial markers characterising the AAF process and develop a microbiota-based strategy to monitor AAF process. Moreover, the correlation between the microbial succession and environmental factors showed that the gradient of titratable acidity was the most important driver to promote succession of microbiota (Fig. 2c). Elevated levels of AA and LA in vinegar Pei resulted in a specially acidic stress, which selected most of acid-tolerant microbes such as Ga. europaeus29. Another important factor was the alcohol stress, which is a preferred carbon source for growth of functional microbes during vinegar production36. Eventually, a well-balanced and robust community is formed via long-time environmental selection.It is interesting that the grouping of AAF process based on microbial assembly (day 0, days 1–9 and days 10–18) is basically accordance with the grouping based on flavours (day 0, days 1–7 and days 8–18) (Figs 2 and 3b), suggesting the uniformity and high correlation between the evolution of microbiota and the change of flavours. There are few investigations of the correlation between microbiota and flavours in traditional fermented foods6. Here, O2PLS approach was used to integrate the microbiota dataset and flavours dataset in order to dig into the association between microbiota and flavours in vinegar Pei during MSSF process. In this study, more bacterial genera showed a higher correlation with three flavour sets (|ρ| > 0.8) than fungal genera (Fig. 4), which indicated bacterial community might be the main producer for vinegar flavours. Seven genera including Acetobacter, Lactobacillus, Enhydrobacter, Lactococcus, Gluconacetobacer, Bacillus and Staphylococcus were selected as functional core microbiota for AAF of Zhenjiang aromatic vinegar. Moreover, the function of core microbiota predicted by PICRUSt analysis accounted for more than 80% of functions of the microbial community in vinegar Pei. Among seven functional genera, Acetobacter and Lactobacillus are major functional microbes that have been studied extensively in vinegar industry4,29. In the further work, A. pasteurianus, a main species isolated from vinegar Pei, was added at the beginning of AAF to augment the flavours production of Zhenjiang aromatic vinegar. The result showed that the temperature of vinegar Pei in AAF process augmented with A. pasteurianus increased faster than that in the non-augmented AAF process (control) (Fig. S8a). The level of total acids in the AAF process of adding A. pasteurianus was higher than control and the level of total acids at the 15th day of AAF process of adding A. pasteurianus was equivalent to the total acids at the 18th day of control AAF process (Fig. S8b), which suggested the fermented period might be shortened by adding A. pasteurianus. Moreover, variety of flavours such as AA, Glu, 2,3-butanediol and ligustrazine (Table S11) were increased at the end of augmented AAF process compared with the control, which partly validated the correlation between Acetobacter and flavours. The structure and dynamics of microbiota after adding A. pasteurianus are being studied. According to our knowledge, this is the first report to systemic analyse the relationship between the structure (genotype) and function (phenotype) of microbial community in traditional fermented foods.MethodsStudy design and samplingThe framework of the experiment design was shown in Fig. S1a. The AAF of Zhenjiang aromatic vinegar was carried out from July to October, 2014 in Jiangsu Hengshun Vinegar Industry Co., Ltd., China. Vinegar Pei from three randomly selected AAF batches (denoted as #5, #8 and #9) were sampled every day using a sterilized cylinder-shaped sampler (Puluody, Xi’an, China). Meanwhile, the alcohol mash (#v_am), starter Pei (#v_sp) and a mixture of raw materials (#v_mp) including alcohol mash, wheat bran, chaff and starter Pei were collected. In order to obtain the most unbiased samples, vinegar Pei at the four vertexes and the centre of the pool were collected from top to bottom, mixed thoroughly and then reduced by coning and quartering repeatedly (Fig. S1c). About 500 g of sample was sealed in a sterile plastic bag and stored at −20 °C before further analysis. During the AAF process, the temperature and moisture content of vinegar Pei were 34–46 °C and 60–70%, respectively. The AAF lasted 18 days and a total of 62 fresh samples were obtained for further analysis. Detailed information of samples is shown in Dataset S1.DNA extraction, amplicon and sequencingDNA extraction using the CTAB-based method was applied in this study37. For bacteria, the V4–V5 domains of 16S rRNA genes were amplified using primers 515F and 907R38. For fungi, the internal transcribed spacer (ITS) region were amplified with primers 1737F and 2043R39. The sequences of primers are listed in Table S12. Amplicons were submitted to the Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China) for illumina paired-end library preparation, cluster generation and 300-bp paired-end sequencing on a MiSeq instrument in two separate runs. The run of bacterial 16S rRNA generated 1,257,819 reads (396.42 nt mean length) and the run of fungal ITS generated 1,224,296 reads (263.41 nt mean length). Details of the DNA extraction and PCR amplification are described in Supplemental information.Microbial biomass analysis by quantitative real-time PCRTo estimate the biomass of bacteria and fungi during the AAF process of Zhenjiang aromatic vinegar, qRT-PCR was performed using a CFX connect Real-Time system (Bio-Rad, California, US) with commercial kit (SYBR Premix Ex Taq, Takara, Dalian, China). The total genomic DNA from Pei was measured (Nanodrop 2000, Wilmington, US) and used as the template to amplify bacteria using primers40 340F and 758R and fungi using primers5 Y1 and Y2. The specificity of amplification was determined by melting curve analysis. For determination of the number of bacterial and fungal amount in each sample, fluorescent signals, detected from 10 times serial dilution (from 10E + 14 copies/μL to 10E + 3 copies/μL) in the linear range of the assay, were averaged and compared to a standard curve generated with standard DNA in the same experiment41. The sequences of primers are listed in Table S12. Details of PCR amplification are described in Supplemental information.Sequence processing and community structure analysisRaw reads were de-multiplexed, quality-filtered and analysed using QIIME (v.1.17)42. The representative OTU sequences were annotated using the RDP bacterial 16S rRNA database (Release 11.1) and the UNITE fungal ITS database (Release 6.0)43 by a QIIME-based wrapper of RDP-classifier (v.2.2)44. Alpha-diversity and β-diversity estimates were calculated using hellinger distance between samples for bacterial 16S rRNA reads and fungal ITS reads with 97% identity. Principal component were computed from the resulting distance matrices to compress dimensionality and visualise the relationships between samples according to PCA plots45. To determine whether sample classifications (different fermentation phase) contained differences in phylogenetic or species diversity, analysis of molecular variance (AMOVA)46 was used to test significant differences between sample groups based on hellinger distance matrices. Metastats was used to determine which taxa resulted in these differences between sample groups47. Moreover, environmental conditions do correlate with variation in community composition; spearman correlation was applied to explore the potential determiner for the succession of bacterial and fungal community (the first principal component) in vinegar Pei. Details of the sequence processing and statistical analyses are summarized in Supplemental information.Flavours analysis and multivariate data analysisThe contents of fructose, glucose, OAs, AAs and VFs were detected by chromatography. PCA and HCA were used to investigate the flavours data during AAF process. In HCA, the distance between observations was calculated using Ward’s method. In PCA, we superimposed the score vectors and loading vectors based on the correlation scaling method, leading to the new vectors t(corr) and p(corr). Then, the new vectors of the first two components were visualized by a biplot. According to the relative positions between observations and variables, we were able to determine which flavours were highly correlated with each AAF group. Before analysis, the flavours data were normalised using the min-max method. PCA and HCA were performed in SIMCA 14 (demo v.1.0.1) (Umetrics AB, Umeå, Sweden). Details of flavours analysis are summarized in Supplemental information.Correlation analysis between microbiota and flavours during AAF processAs for microbiota in Pei, the top 100 bacterial genera and top 100 fungal genera were further analysed according to rank of sum of abundance. For flavours, total 88 flavours including 2 sugars, 9 OAs, 18 AAs and 59 VFs were applied to investigate the relationship with microbiota. O2PLS modelling was used to unveil the association between microbiota at genus level and each flavour during AAF, in which, microbiota data for 200 genera (defined as X matrix) were mapped to flavours data (defined as Y matrix)16. O2PLS method consists of simultaneous projection of both the X and Y matrices on low dimensional hyper planes13. The number of components in respective set of O2PLS model is evaluated by seven-fold cross-validation. Variable Importance in the Projection (VIP) and a pair-wise correlation matrix (|ρ| > 0.7) were employed to identify potential functional microbiota in vinegar Pei. Terms with larger VIP value (>1), are the most relevant for explaining Y variables. The correlation matrix shows the pair-wise correlation between all variables (X and Y), in which the value of correlation coefficient represents the extent of the linear association between the two terms, ranging from −1 to 1. O2PLS analysis was performed using the SIMCA 14 (demo v.1.0.1) (Umetrics AB, Umeå, Sweden). Further statistic analyses and graphics were performed in Microsoft® Excel and R software (v.2.14.1). The correlation between microbiota and flavours was visualised via Cytoscape (v.2.8.3). Details for correlation analysis are listed in Supplemental information.Predicted function of the core microbiota and non-core microbiota in vinegar PeiTo validate the function of the core microbiota for the whole community in vinegar Pei, phylogenetic investigation of communities by reconstruction of unobserved states (PICRUSt), was used to predict which gene families were present48. Given that the copy number and function of bacteria were more abundant than fungi, PICRUSt was performed base on the bacterial 16S gene surveys. For this analysis, OTUs were closed-reference picked against the Greengenes by QIIME (v.1.7). The functional core taxonomies were filtered as a separate dataset of core microbiota while the remaining taxonomies were regarded as another dataset of non-core microbiota. The two datasets were normalised, predicted and categorised according to online protocols of PICRUSt (http://huttenhower.sph.harvard.edu/galaxy). The predicted functions of the core microbiota and non-core microbiota were compared and visualised in Microsoft® Excel and Origin (v.8.0).Additional InformationAccession codes: The sequences data reported in this paper have been deposited in the GenBank database (No. SRP059163).How to cite this article: Wang, Z.-M. et al. Exploring flavour-producing core microbiota in multispecies solid-state fermentation of traditional Chinese vinegar. Sci. Rep. 6, 26818; doi: 10.1038/srep26818 (2016).

ReferencesHugenholtz, J. Traditional biotechnology for new foods and beverages. Curr. Opin. Biotechnol. 24, 155–159 (2013).CAS

PubMed

Google Scholar

Blandino, A., Al-Aseeri, M. E., Pandiella, S. S., Cantero, D. & Webb, C. Cereal-based fermented foods and beverages. Food Res. Int. 36, 527–543 (2003).CAS

Google Scholar

Wang, H. Y., Gao, Y. B., Fan, Q. W. & Xu, Y. Characterization and comparison of microbial community of different typical Chinese liquor Daqus by PCR-DGGE. Lett. Appl. Microbiol. 53, 134–140 (2011).CAS

PubMed

Google Scholar

Wu, J. J., Ma, Y. K., Zhang, F. F. & Chen, F. S. Biodiversity of yeasts, lactic acid bacteria and acetic acid bacteria in the fermentation of “Shanxi aged vinegar”, a traditional Chinese vinegar. Food Microbiol. 30, 289–297 (2012).CAS

PubMed

Google Scholar

Xu, W. et al. Monitoring the microbial community during solid-state acetic acid fermentation of Zhenjiang aromatic vinegar. Food Microbiol. 28, 1175–1181 (2011).CAS

PubMed

Google Scholar

Jung, J. Y. et al. Metagenomic analysis of kimchi, a traditional Korean fermented food. Appl. Environ. Microbiol. 77, 2264–2274 (2011).CAS

PubMed

PubMed Central

Google Scholar

Randazzo, C. L., Heilig, H., Restuccia, C., Giudici, P. & Caggia, C. Bacterial population in traditional sourdough evaluated by molecular methods. J. Appl. Microbiol. 99, 251–258 (2005).CAS

PubMed

Google Scholar

Weckx, S. et al. Metatranscriptome analysis for insight into whole-ecosystem gene expression during spontaneous wheat and spelt sourdough fermentations. Appl. Environ. Microbiol. 77, 618–626 (2011).CAS

PubMed

Google Scholar

Jung, J. Y. et al. Metatranscriptomic analysis of lactic acid bacterial gene expression during kimchi fermentation. Int. J. Food Microbiol. 163, 171–179 (2013).CAS

PubMed

Google Scholar

Bisanz, J. E., Macklaim, J. M., Gloor, G. B. & Reid, G. Bacterial metatranscriptome analysis of a probiotic yogurt using an RNA-Seq approach. Int. Dairy J. 39, 284–292 (2014).CAS

Google Scholar

Solieri, L., Dakal, T. C. & Giudici, P. Next-generation sequencing and its potential impact on food microbial genomics. Ann. Microbiol. 63, 21–37 (2013).CAS

Google Scholar

Delmont, T. O. et al. Metagenomic mining for microbiologists. ISME J. 5, 1837–1843 (2011).CAS

PubMed

PubMed Central

Google Scholar

Trygg, J. O2-PLS for qualitative and quantitative analysis in multivariate calibration. J. Chemometr. 16, 283–293 (2002).CAS

Google Scholar

Rantalainen, M. et al. Statistically integrated metabonomic–proteomic studies on a human prostate cancer xenograft model in mice. J. Proteome Res. 5, 2642–2655 (2006).CAS

PubMed

Google Scholar

Li, M. et al. Symbiotic gut microbes modulate human metabolic phenotypes. P. Natl. Acad. Sci. 105, 2117–2122 (2008).ADS

CAS

Google Scholar

Bylesjö, M., Eriksson, D., Kusano, M., Moritz, T. & Trygg, J. Data integration in plant biology: the O2PLS method for combined modeling of transcript and metabolite data. Plant J. 52, 1181–1191 (2007).PubMed

Google Scholar

Wang, Z. M. et al. Batch-to-batch uniformity of bacterial community succession and flavor formation in the fermentation of Zhenjiang aromatic vinegar. Food Microbiol. 50, 64–69 (2015).CAS

PubMed

Google Scholar

Papotti, G. et al. Traditional balsamic vinegar and balsamic vinegar of Modena analyzed by nuclear magnetic resonance spectroscopy coupled with multivariate data analysis. LWT - Food Sci. Technol. 60, 1017–1024 (2015).CAS

Google Scholar

Jung, J. Y., Lee, S. H., Lee, H. J. & Jeon, C. O. Microbial succession and metabolite changes during fermentation of saeu-jeot: traditional Korean salted seafood. Food Microbiol. 34, 360–368 (2013).CAS

PubMed

Google Scholar

Anupam, G. & Toshiaki, O. Dynamics of aroma-active volatiles in miso prepared from lizardfish meat and soy during fermentation: a comparative analysis. Int. J. Nutr. Food Sci. 1, 1–12 (2012).

Google Scholar

Uysal, R. S., Soykut, E. A., Boyaci, I. H. & Topcu, A. Monitoring multiple components in vinegar fermentation using Raman spectroscopy. Food Chem. 141, 4333–4343 (2013).CAS

PubMed

Google Scholar

Feng, Y. Z. et al. Changes in fatty acid composition and lipid profile during koji fermentation and their relationships with soy sauce flavor. Food Chem. 158, 438–444 (2014).CAS

PubMed

Google Scholar

Jo, Y. et al. Physicochemical properties and volatile components of wine vinegars with high acidity based on fermentation stage and initial alcohol concentration. Food Sci. Biotechnol. 24, 445–452 (2015).CAS

Google Scholar

Yoshimura, M. et al. Antihypertensive effect of a γ-aminobutyric acid rich tomato cultivar ‘DG03-9’ in spontaneously hypertensive rats. J. Agric. Food Chem. 58, 615–619 (2010).CAS

PubMed

Google Scholar

Wu, J. J., Gullo, M., Chen, F. S. & Giudici, P. Diversity of Acetobacter pasteurianus strains isolated from solid-state fermentation of cereal vinegars. Curr. Microbiol. 60, 280–286 (2010).CAS

PubMed

Google Scholar

Nie, Z. Q. et al. Exploring microbial succession and diversity during solid-state fermentation of Tianjin duliu mature vinegar. Bioresour. Technol. 148, 325–333 (2013).CAS

PubMed

Google Scholar

Nie, Z. Q., Zheng, Y., Du, H., Xie, S. & Wang, M. Dynamics and diversity of microbial community succession in traditional fermentation of Shanxi aged vinegar. Food Microbiol. 47, 62–68 (2015).CAS

PubMed

Google Scholar

Nanda, K. et al. Characterization of acetic acid bacteria in traditional acetic acid fermentation of rice vinegar (Komesu) and unpolished rice vinegar (Kurosu) produced in Japan. Appl. Environ. Microbiol. 67, 986–990 (2001).CAS

PubMed

PubMed Central

Google Scholar

Gullo, M., De Vero, L. & Giudici, P. Succession of selected strains of Acetobacter pasteurianus and other acetic acid bacteria in traditional balsamic vinegar. Appl. Environ. Microbiol. 75, 2585–2589 (2009).CAS

PubMed

PubMed Central

Google Scholar

De Vero, L. et al. Application of denaturing gradient gel electrophoresis (DGGE) analysis to evaluate acetic acid bacteria in traditional balsamic vinegar. Food Microbiol. 23, 809–813 (2006).CAS

PubMed

Google Scholar

Gullo, M., Caggia, C., De Vero, L. & Giudici, P. Characterization of acetic acid bacteria in “traditional balsamic vinegar”. Int. J. Food Microbiol. 106, 209–212 (2006).CAS

PubMed

Google Scholar

Haruta, S. et al. Succession of bacterial and fungal communities during a traditional pot fermentation of rice vinegar assessed by PCR-mediated denaturing gradient gel electrophoresis. Int. J. Food Microbiol. 109, 79–87 (2006).CAS

PubMed

Google Scholar

Solieri, L., Landi, S., De Vero, L. & Giudici, P. Molecular assessment of indigenous yeast population from traditional balsamic vinegar. J. Appl. Microbiol. 101, 63–71 (2006).CAS

PubMed

Google Scholar

Charpentier, C. et al. Release of nucleotides and nucleosides during yeast autolysis: kinetics and potential impact on flavor. J. Agric. Food Chem. 53, 3000–3007 (2005).CAS

PubMed

Google Scholar

Alexandre, H. & Guilloux-Benatier, M. Yeast autolysis in sparkling wine - a review. Aust. J. Grape Wine Res. 12, 119–127 (2006).CAS

Google Scholar

Gullo, M. & Giudici, P. Acetic acid bacteria in traditional balsamic vinegar: phenotypic traits relevant for starter cultures selection. Int. J. Food Microbiol. 125, 46–53 (2008).CAS

PubMed

Google Scholar

Zhou, J. Z., Bruns, M. A. & Tiedje, J. M. DNA recovery from soils of diverse composition. Appl. Environ. Microbiol. 62, 316–322 (1996).CAS

PubMed

PubMed Central

Google Scholar

Xiong, J. et al. Geographic distance and pH drive bacterial distribution in alkaline lake sediments across Tibetan Plateau. Environ. Microbiol. 14, 2457–2466 (2012).CAS

PubMed

PubMed Central

Google Scholar

Bellemain, E. et al. ITS as an environmental DNA barcode for fungi: an in silico approach reveals potential PCR biases. BMC Microbiol. 10, 189 (2010).PubMed

PubMed Central

Google Scholar

Juck, D., Charles, T., Whyte, L. G. & Greer, C. W. Polyphasic microbial community analysis of petroleum hydrocarbon-contaminated soils from two northern Canadian communities. FEMS Microbiol. Ecol. 33, 241–249 (2000).CAS

PubMed

Google Scholar

Park, E. J. et al. Application of quantitative real-time PCR for enumeration of total bacterial, archaeal and yeast populations in kimchi. J. Microbiol. 47, 682–685 (2009).CAS

PubMed

Google Scholar

Caporaso, J. G. et al. QIIME allows analysis of high-throughput community sequencing data. Nat. Methods 7, 335–336 (2010).CAS

PubMed

PubMed Central

Google Scholar

Abarenkov, K. et al. The UNITE database for molecular identification of fungi – recent updates and future perspectives. New Phytol. 186, 281–285 (2010).PubMed

Google Scholar

Wang, Q., Garrity, G. M., Tiedje, J. M. & Cole, J. R. Naïve bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Appl. Environ. Microbiol. 73, 5261–5267 (2007).CAS

PubMed

PubMed Central

Google Scholar

Wang, Y. et al. Comparison of the levels of bacterial diversity in freshwater, intertidal wetland and marine sediments by using millions of illumina tags. Appl. Environ. Microbiol. 78, 8264–8271 (2012).CAS

PubMed

PubMed Central

Google Scholar

Meirmans, P. G. Using the AMOVA framwork to estimate a standardized genetic differential measure. Evolution 60, 2399–2402 (2006).PubMed

Google Scholar

White, J. R., Nagarajan, N. & Pop, M. Statistical methods for detecting differentially abundant features in clinical metagenomic samples. Plos Comput. Biol. 5, e1000352 (2009).ADS

PubMed

PubMed Central

Google Scholar

Langille, M. G. I. et al. Predictive functional profiling of microbial communities using 16S rRNA marker gene sequences. Nat. Biotechnol. 31, 814–821 (2013).CAS

PubMed

PubMed Central

Google Scholar

Download referencesAcknowledgementsThis work was supported by two grants from the National Nature Science Foundation of China (No. 31271922 and No. 31530055), three grants from the High Tech Development Program of China (863 Project) (No. 2012AA021301, No. 2013AA102106 and No. 2014AA021501) and two grants from the National Engineering Research Centre of Solid-State Brewing (No. 2011B2211 and No. GCKF201109).Author informationAuthor notesWang Zong-Min and Lu Zhen-Ming contributed equally to this work.Authors and AffiliationsSchool of Pharmaceutical Science, Key Laboratory of Industrial Biotechnology of Ministry of Education, Jiangnan University, Wuxi, 214122, ChinaZong-Min Wang, Zhen-Ming Lu, Jin-Song Shi & Zheng-Hong XuTianjin Key Laboratory for Industrial Biological Systems and Bioprocessing Engineering, Tianjin Institute of Industrial Biotechnology, Chinese Academy of Sciences, Tianjin, 300308, ChinaZhen-Ming Lu & Zheng-Hong XuNational Engineering Research Centre of Solid-State Brewing, Luzhou, 646000, ChinaJin-Song Shi & Zheng-Hong XuAuthorsZong-Min WangView author publicationsYou can also search for this author in

PubMed Google ScholarZhen-Ming LuView author publicationsYou can also search for this author in

PubMed Google ScholarJin-Song ShiView author publicationsYou can also search for this author in

PubMed Google ScholarZheng-Hong XuView author publicationsYou can also search for this author in

PubMed Google ScholarContributionsZ.-H.X. and J.-S.S. conceived and designed the experiments. Z.-M.W. and Z.-M.L. performed the experiments, analysed the data and wrote the paper. All authors reviewed the manuscript.Ethics declarations

Competing interests

The authors declare no competing financial interests.

Electronic supplementary materialSupplementary InformationSupplementary Dataset 1Supplementary Dataset 2Rights and permissions

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Reprints and permissionsAbout this articleCite this articleWang, ZM., Lu, ZM., Shi, JS. et al. Exploring flavour-producing core microbiota in multispecies solid-state fermentation of traditional Chinese vinegar.

Sci Rep 6, 26818 (2016). https://doi.org/10.1038/srep26818Download citationReceived: 17 December 2015Accepted: 18 March 2016Published: 31 May 2016DOI: https://doi.org/10.1038/srep26818Share this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard

Provided by the Springer Nature SharedIt content-sharing initiative

This article is cited by

Exploring the correlation of metabolites changes and microbial succession in solid-state fermentation of Sichuan Sun-dried vinegar

Ke DongWeizhou LiXiaofang Pei

BMC Microbiology (2023)

The bacterial succession and its role in flavor compounds formation during the fermentation of cigar tobacco leaves

Hongyang SiKun ZhouMingqin Zhao

Bioresources and Bioprocessing (2023)

Processing Technologies and Flavor Analysis of Chinese Cereal Vinegar: a Comprehensive Review

Sam Al-DalaliFuping ZhengBaoguo Sun

Food Analytical Methods (2023)

Profiling the role of microorganisms in quality improvement of the aged flue-cured tobacco

Xinying WuWen CaiJuan Zhang

BMC Microbiology (2022)

Microbial succession and exploration of higher alcohols-producing core bacteria in northern Huangjiu fermentation

Yi YanLeping SunQing Ren

AMB Express (2022)

CommentsBy submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Download PDF

Explore content

Research articles

News & Comment

Collections

Subjects

RSS feed

About the journal

Open Access Fees and Funding

About Scientific Reports

Contact

Journal policies

Calls for Papers

Guide to referees

Editor's Choice

Journal highlights

Publish with us

For authors

Language editing services

Submit manuscript

Search articles by subject, keyword or author

Show results from

All journals

This journal

Advanced search

Quick links

Explore articles by subject

Find a job

Guide to authors

Editorial policies

Scientific Reports (Sci Rep)

ISSN 2045-2322 (online)

nature.com sitemap

About Nature Portfolio

About us

Press releases

Press office

Discover content

Journals A-Z

Articles by subject

Protocol Exchange

Nature Index

Publishing policies

Nature portfolio policies

Open access

Author & Researcher services

Reprints & permissions

Research data

Language editing

Scientific editing

Nature Masterclasses

Research Solutions

Libraries & institutions

Librarian service & tools

Librarian portal

Open research

Recommend to library

Advertising & partnerships

Advertising

Partnerships & Services

Media kits

Branded

content

Professional development

Nature Careers

Nature

Conferences

Regional websites

Nature Africa

Nature China

Nature India

Nature Italy

Nature Japan

Nature Korea

Nature Middle East

Privacy

Policy

Use

of cookies

Your privacy choices/Manage cookies

Legal

notice

Accessibility

statement

Terms & Conditions

Your US state privacy rights

Close banner

Email address

I agree my information will be processed in accordance with the Nature and Springer Nature Limited Privacy Policy.

Close banner

Get the most important science stories of the day, free in your inbox.

Follow

tp钱包中国官方网站|mssf

tp钱包中国官方网站|mssf

Maruti Suzuki Smart Finance - Car Loans for Alto, Wagon R, Swift, Dzire, Ertiga, Vitara Brezza, S-PRESSO, Celerio, Eeco

搜狗皮肤编辑器 for Mac

问下官方皮肤是 .ssf 的 这个.mssf要怎么安装 双击没有应用程序能打开 · Issue #1 · xiaochunjimmy/Sogou-Input-Skin · GitHub

Sogou-Input-Skin: 搜狗输入法皮肤

P3 | msater-slave sampling filter (MSSF) (JSSC-2016-03)_a 2.4-ghz 6.4-mw fractional-n inductorless rf synt-CSDN博客

Page restricted | ScienceDirect

Remote Sensing | Free Full-Text | MSSF: A Novel Mutual Structure Shift Feature for Removing Incorrect Keypoint Correspondences between Images

GitHub - fkxxyz/ssfconv: Sogou input method skin file (.ssf file) converter, supports conversion to fcitx or fcitx5 format.

MSSF-GCN: Multi-scale Structural and Semantic Information Fusion Graph Convolutional Network for Controversy Detection | SpringerLink

Exploring flavour-producing core microbiota in multispecies solid-state fermentation of traditional Chinese vinegar | Scientific Reports

最近的新闻

您可能喜欢的文章

Imtoken钱包2.9.7|mpi

比特派app官方下载网址|mst

比特派点对点交易

比特派 使用方法

比特派怎么重置wifi

比特派 冷钱包 手表

比特派不用登录密码

问下官方皮肤是 .ssf 的这个.mssf要怎么安装双击没有应用程序能打开 · Issue #1 · xiaochunjimmy/Sogou-Input-Skin · GitHub

比特派使用方法

比特派冷钱包手表