您现在的位置是:首页 >学无止境 >目标检测论文总结网站首页学无止境

目标检测论文总结

传参员小林 2024-06-15 12:01:02
简介目标检测论文总结

1.目标检测论文

EI https://www.engineeringvillage.com/search/quick.url

其他

A YOLOv3-based Deep Learning Application Research for
Condition Monitoring of Rail Thermite Welded Joints          
 
https://dl.acm.org/doi/10.1145/3388818.3388827         

Traffic Sign(s) DetectionRecognition

1

CCTSDB 2021: A More Comprehensive Traffic Sign Detection Benchmark
2区,这论文在别人旧版本上增加采集了4000张图片数据集,用九种模型跑分(6种评估维度)

http://hcisj.com/articles/?HCIS202212023

2

Traffic Sign Detection via Improved Sparse R-CNN for Autonomous Vehicles
4区,提出了新的数据集,收集数据的装备看起来很唬人,可学习其写作手法

https://www.engineeringvillage.com/app/doc/?docid=cpx_5af19ccd17fd7bd6ee2M61b91017816328&pageSize=25&index=17&searchId=9301f2b5f9544b31a3d8ea95cbc592ce&resultsCount=3429&usageZone=resultslist&usageOrigin=searchresults&searchType=Quick

3

Long-Tailed Traffic Sign Detection Using Attentive
Fusion and Hierarchical Group Softmax
IEEE,引用了公开数据集,IF=9;IEEE Transactions on Intelligent Transportation Systems

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9870750&tag=1

4

Traffic signal image detection technology based on YOLO
21年OA;YOLO34对比;数据来源于TT100K和视频采集得100张图,处理后总共1w张图,100多个类别(我怀疑!)
https://www.engineeringvillage.com/app/doc/?docid=cpx_M816368317b07fe17dcM7d4c1017816328&pageSize=25&index=2&searchId=bf793336369d4197a4062e6d0f6415e2&resultsCount=4&usageZone=resultslist&usageOrigin=searchresults&searchType=Expert

5

Traffic Signs Detection and Recognition System using Deep Learning
19年OA;1000张德国数据图4个分类;说是多个算法结合

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9014763

6

Traffic Sign Recognition with a small convolutional neural network
19OA;德国和比利时数据集;CNN和小改的CNN

https://www.engineeringvillage.com/app/doc/?docid=cpx_406f2ac917251502fa6M7e9d10178163190&pageSize=25&index=9&searchId=25fc2ffaf3a64c05b7f5b14581106b05&resultsCount=204&usageZone=resultslist&usageOrigin=searchresults&searchType=Expert

7

Application of New Generation Artificial Intelligence in Traffic Informatization
21OA;纯水;算法都不改
https://iopscience.iop.org/article/10.1088/1742-6596/1881/2/022070        

8

A SECI Method Based on Improved YOLOv4 for Traffic Sign Detection and Recognition
22OA;改进YOLO;抽取部分TT100K数据集
https://iopscience.iop.org/article/10.1088/1742-6596/2337/1/012001           

9

Traffic Sign Detection and Recognition for Autonomous Driving in Virtual Simulation Environment


22OA;数据集源于比赛;RetinaNet

https://www.engineeringvillage.com/app/doc/?docid=cpx_187a1b8183c37112e4M7fb91017816355&pageSize=25&index=4&searchId=9e58187458474fbaa5730766519550e0&resultsCount=204&usageZone=resultslist&usageOrigin=searchresults&searchType=Expert

10

TSR-YOLO: A Chinese Traffic Sign Recognition Algorithm for Intelligent Vehicles in Complex Scenes
23OA,JA;;改进YOLO;CCTSDB 2021
https://www.mdpi.com/1424-8220/23/2/749

11

L-YOLO:适用于车载边缘计算的实时交通标识检测模型
20北大核心;基于Tiny YOLO改进
https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C44YLTlOAiTRKibYlV5Vjs7iy_Rpms2pqwbFRRUtoUImHYcwLJG1DVerIKTwqecquIaQF0UHe7a5TcjHaMFU4nWV&uniplatform=NZKPT

12

基于特征融合级联网络的交通标识牌检测算法
20北核心;算法在Faster R-CNN框架的基础上增加了特征融合模块,Tsinghua-Tencent 100k
https://kns.cnki.net/KXReader/Detail?invoice=uG44WynIFA50KI1fCM0%2FqR96eyYDgRLWY86t10W5Y0M7wh7rUdYBPvR3EVl9HEJU4LL%2FOVNWAzKHLS6IUSlAs8iLud%2F5fMeg5OjieHzq0M%2F1%2Bu9WiqXXUkYLfuML1XMjlhSXdMbG9lJkJbLLZHt%2BLtl8Pu%2BsCebWhPuO7e2uIvM%3D&DBCODE=CJFD&FileName=JYRJ202204030&TABLEName=cjfdlast2022&nonce=957B6E8099AB42B2802FC84E87F861D0&TIMESTAMP=1682408268215&uid=

13

复杂环境下的交通标志检测与识别方法综述
编写论文可以参考
https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C44YLTlOAiTRKibYlV5Vjs7iy_Rpms2pqwbFRRUtoUImHThQDoN4QwaAQlg3DC6G4qHVrfcGkbQua8HXJc67vm0X&uniplatform=NZKPT

14

在这里插入图片描述

基于YOLOv5算法的交通标志识别技术研究
21北大核心;改进YOLOv5(EIOU损失函数;加权Cluster非极大值抑制NMS);长沙理工大学制作的CCTSDB交通标志数据集上训练的模型的mAP值达到了84.35%,比原始的YOLOv5算法提高了6.23%。

https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C44YLTlOAiTRKibYlV5Vjs7iJTKGjg9uTdeTsOI_ra5_XYNPqXQKnzyOCl1vVpgvK7C1Ng5fDAvJQejSbWuYpmem&uniplatform=NZKPT

15

在这里插入图片描述

基于改进YOLOv5s的交通标志识别算法
22北核;改进YOLOv5s(引入MobileNetv3主干网络;在特征融合中采用AFF模块;采用Matrix NMS筛选候选框);CCTSDB

https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C44YLTlOAiTRKibYlV5Vjs7iJTKGjg9uTdeTsOI_ra5_XU69SyxAE7HoELIggZUwcrWaiInRvJoBrQkLVnvokAzL&uniplatform=NZKPT## 

16

一种改进YOLOv4的交通标志识别算法
22北核;yolov4(深度可分离卷积;BiFPN;Focal损失函数代);多种模块对比
https://kns.cnki.net/kcms2/article/abstract?v=Il_FVXPSRQ-RBnmV8io-ko8LX1TQFeEiNEHfaWL35oQIcw33JOC8gzk8PNMQ5Yff71MiBGnGeIIJL0CHn4Z_4BfLV-Gth4Hbc9-ULO10v10=&uniplatform=NZKPT&language=CHS

17

在这里插入图片描述

改进YOLOv5s的交通标志识别算法
22北核;改进YOLOv5s(copy-paste进行数据增强;引入Ghost来构建网络;坐标注意力机制(coordinate attention))
https://kns.cnki.net/KXReader/Detail?invoice=Rm2DTQWkXKoSKzOMFrMFAUmrdOrMnpjsPHV%2FVq6ZLK6X0l0vRP3CUQfLLxa9h0clECXvWLi4V52GLCHzqwFngdHhuiSGv7Lp9PhbT5ise%2FRkomJ9oMpv8ET%2BdxocWjvoaK57zpm3bosBJS%2F8FtlvDULPZkG8nC%2BdhOi6udd28TM%3D&DBCODE=CJFD&FileName=XTYY202212029&TABLEName=cjfdlast2023&nonce=BE36EDE3C0784372B043AEC063F8C063&TIMESTAMP=1682927986778&uid=

18

在这里插入图片描述

一种基于改进YOLOv5s-Ghost网络的交通标志识别方法
23北核;改进的YOLOv5s-Ghost(3×3运算核Ghost Net;Ghost Bottleneck CSP;)
https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C44YLTlOAiTRKibYlV5Vjs7ioT0BO4yQ4m_mOgeS2ml3UOz4coc21iG0f3XNJDow5A4hR_VD9HpNRwzS8AJkgXKC&uniplatform=NZKPT

19

在这里插入图片描述

基于SA-YOLOv5的交通标志目标检测研究
23北;基于SA-YOLOv5(Shuffle注意力模块;CBAM)
https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C45S0n9fL2suRadTyEVl2pW9UrhTDCdPD66mU_P1gXq-hp0xHEWziSMECUdb-QG2fVGktMIlbUvsW9jWhkMlHMEa&uniplatform=NZKPT

20

在这里插入图片描述

基于YOLOv5的雾霾天气下交通标志识别
23北;(融入卷积注意力机制,在空间维度和通道维度上进行特征增强;BiFPN作为neck层;CIoU;K-means聚类算法在TT100K和CODA数据集);)TT100K数据集中缺少雾霾天气下的图像,本文采用python的第三方库pillow对数据集进行雾霾增强;
https://kns.cnki.net/KXReader/Detail?invoice=WsS2TYM%2BvuIFSIt7hAkl7XJ5UgXQnlppXPCBtcRh%2F3PK3XBUwudZQWZmE32KN8c7NWMJKJzKrsxB9wkisPsR%2BbVGUXFvLUsX2F6tIIChoMVUpvzudmDgyGtqQChCjDv5%2BXMrq985ZzTxCDHw2o4LlRKAu3rE%2BFyVxPPNsgQicdM%3D&DBCODE=CAPJ&FileName=DZCL20230220006&TABLEName=capjlast&nonce=1CDCEA419FB345C79D04D8C29A9DBF87&TIMESTAMP=1682935210318&uid=

21

基于坐标注意力的轻量级交通标志识别模型
23bei;YOLOv5(坐标注意力(CA);特征融合网络中加入跨层连接;改进的CIoU函数)
https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C44YLTlOAiTRKibYlV5Vjs7ioT0BO4yQ4m_mOgeS2ml3UBV8gSVxETk7-d2mGbO08hjeILhgFY9xelSoSjam7gDz&uniplatform=NZKPT

22

基于YOLOv5l和ViT的交通标志检测识别方法
https://kns.cnki.net/KXReader/Detail?invoice=O8szC%2FN5Q4VbsJspvxVnaWONoJVpziQyE4sdP2whOKnGftAR8jlIZ1xd3EX1NhkFET41Co03nhT%2FQ8lV2TUMiDmlsVvsMw%2Bko7xs8pEkBg52PrCfyO%2Bf1Xah3H0ICfSALc5J9dJLhjir%2ByoaLrbhWedSffO8ITYT%2F4MTFd2P404%3D&DBCODE=CJFD&FileName=KXJS202227035&TABLEName=cjfdlast2022&nonce=F06C1A4EC924432DAAEB53EF37589D54&TIMESTAMP=1682940692920&uid=

23

基于YOLOv5-EA的交通标志识别
22非;YOLOv5-EA(有效通道注意力机制;通过增加小尺度检测层;BSConv代替了正则卷积);TT100K
https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C44YLTlOAiTRKibYlV5Vjs7iJTKGjg9uTdeTsOI_ra5_XajyyJSlFizEwHxsaQ-OwbXbREMj40MlIbuCDeG_3jkY&uniplatform=NZKPT

24

一种基于YOLOv5改进的雨天环境交通标志识别检测
22非;YOLOv5(渐进递归网络(PRN)对摄像头采集到的画面进行去雨处理;)
https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C44YLTlOAiTRKibYlV5Vjs7iJTKGjg9uTdeTsOI_ra5_Xf66FhD4vWZrk_F6NPzGDHL8x1nYWtjlO5TJlje965Cp&uniplatform=NZKPT

25

基于注意力机制的交通标志识别
22北;改进YOLOv5(CBAM同时嵌入YOLOv5网络的Backbone和Head部分;改用DIoU Loss)
https://kns.cnki.net/kcms2/article/abstract?v=3uoqIhG8C44YLTlOAiTRKibYlV5Vjs7iJTKGjg9uTdeTsOI_ra5_XT9expSGNIVxo4l6lKJom69I_N8R3Yc2-pfemX84X2qV&uniplatform=NZKPT

26

Traffic Sign Detection Method Using Multi-Color Space Fusion
20ICAICA
https://www.engineeringvillage.com/app/doc/?docid=cpx_M14604e69175423765f3M770210178163190&pageSize=25&index=1&searchId=c2e31097ba804c789a3dce8f3d60619f&resultsCount=2&usageZone=resultslist&usageOrigin=searchresults&searchType=Quick

27

Research on the Application of YOLO v3 in Railway Intruding Objects Recognition
ICAICA 2022
https://www.engineeringvillage.com/app/doc/?docid=cpx_6d6e69e71830f3259f2M648b1017816355&pageSize=25&index=1&searchId=f9b11ea31ce3417e9401d4153ee06436&resultsCount=9&usageZone=resultslist&usageOrigin=searchresults&searchType=Quick

总结改进思路

backbone
neck
head
transformer
注意力机制
2-4个不等创新点
基于YOLOv5的居多
创新点并不是特别复杂
CNN 和 Transformer(ViT)结合的 不少
使用swin、bot等transformer
改进基本上都是在YOLO框架上小改,backbone,neck,head,小幅改进
应用在私有数据集 或者 垂直领域数据集
增加检测层
添加注意力机制(CBAM、SE、SA等)
使用各种卷积模块(eg: Ghostbottleneck)
使用其他loss函数,比如diou giou siou
使用 ResNeSt、densenet、resnet等网络
使用重参数化网络(Repvgg等)
使用各种改进的金字塔池化
一般级别论文基本都是不同模块进行组合、级别高一点的期刊论文 就需要自己改一些特有的结构,有自己的亮点


增加工作量的点:1.采集数据;2.多种算法对比;3.多种模块对比;4.平台部署;


目前思路:1.数据增强;2注意力机制;3换NMS;4抽取部分数据;5K-m
eans;
epoch50-300;16batch;取名可以根据场景取比如雨天

风语者!平时喜欢研究各种技术,目前在从事后端开发工作,热爱生活、热爱工作。