Article(id=1203753463323800184, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1203753457208504777, articleNumber=null, orderNo=null, doi=10.12404/j.issn.1671-1815.2401676, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=research-article, receivedDate=1710086400000, receivedDateStr=2024-03-11, revisedDate=1730649600000, revisedDateStr=2024-11-04, acceptedDate=null, acceptedDateStr=null, onlineDate=1764926790313, onlineDateStr=2025-12-05, pubDate=1737129600000, pubDateStr=2025-01-18, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1764926790313, onlineIssueDateStr=2025-12-05, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1764926790313, creator=13701087609, updateTime=1764926790313, updator=13701087609, issue=Issue{id=1203753457208504777, tenantId=1146029695717560320, journalId=1146123166801305609, year='2025', volume='25', issue='2', pageStart='439', pageEnd='878', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1764926788856, creator=13701087609, updateTime=1764928745558, updator=13701087609, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1203761664261858014, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1203753457208504777, language=EN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1203761664261858015, tenantId=1146029695717560320, journalId=1146123166801305609, issueId=1203753457208504777, language=CN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=683, endPage=694, ext={EN=ArticleExt(id=1203753464607257338, articleId=1203753463323800184, tenantId=1146029695717560320, journalId=1146123166801305609, language=EN, title=A Lightweight Road Disease Detection Algorithm Fused with Deformable Convolution, columnId=1156262729162810294, journalTitle=Science Technology and Engineering, columnName=Papers·Automation and Computational Technology, runingTitle=null, highlight=null, articleAbstract=

In response to the low detection accuracy and high model complexity of existing road damage detection algorithms in complex environments, a lightweight road damage detection algorithm named LDC-YOLOv5 (lightweight deformable convolution YOLOv5) was proposed based on YOLOv5.To address the complexity of real road surface damages, a lightweight feature extraction module was designed using Deformable Conv (deformable convolution) and Depthwise Conv (depthwise convolution) to replace the C3 module in the original network backbone, enabling convolutional kernels to focus on irregular crack damages and enhancing feature extraction for damage detection. To reduce algorithm complexity in the feature fusion stage, a lightweight feature fusion module was constructed using GhostConv to replace the C3 module in the original network neck, lowering network parameters and complexity. Additionally, to prevent missed detections caused by uneven lighting and shadow obstruction, a lightweight attention mechanism, TripletAttention, was introduced in the backbone network to improve the algorithm's understanding of damage information and context. Experiments conducted on the IEEE open dataset RDD2022 and the Kaggle open dataset Road Damage demonstrate that, compared to YOLOv5s, the proposed LDC-YOLOv5 achieves a 1.4% and 4.2% improvement in mAP50 on the two datasets, respectively, with only 67.6% of the model parameters of YOLOv5s.

, correspAuthors=Zi-qiang CHEN, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, authorCompany=null, fund=null, authors=null, authorsList=Ling-xin KONG, Zi-qiang CHEN, Liang-nian JIN, Yan-ying JIANG), CN=ArticleExt(id=1203753468281467055, articleId=1203753463323800184, tenantId=1146029695717560320, journalId=1146123166801305609, language=CN, title=融合可变形卷积的轻量级路面病害检测算法, columnId=1156262729783567290, journalTitle=科学技术与工程, columnName=论文·自动化技术、计算机技术, runingTitle=null, highlight=null, articleAbstract=

针对现有的路面病害检测算法在复杂环境下检测精度低、模型复杂度高的问题,在YOLOv5基础上,提出了一种融合可变形卷积的轻量级路面病害检测算法(lightweight deformable convolution YOLOv5,LDC-YOLOv5)。首先,针对真实路面病害复杂不规整的特点,使用可变形卷积(Deformable Conv)和深度卷积(Depthwise Conv),设计了一种轻量级特征提取模块,代替原网络主干部分的C3模块,使卷积核聚焦在无规则裂缝病害上,增强病害特征提取能力。其次,针对特征融合阶段出现算法复杂度过高的问题,使用轻量级卷积GhostConv,构建一种轻量级特征融合模块,代替原网络颈部网络部分的C3模块,降低网络参数和复杂度;为避免真实路面出现光照不均,出现阴影遮挡路面病害目标而造成的病害漏检的情况,在主干网络部分,引入轻量级注意力机制TripletAttention,增强算法对病害信息上下文之间的理解能力。最后在IEEE公开数据集RDD2022和Kaggle公开数据集Road Damage上进行测试,实验结果表明,与YOLOv5s相比,mAP50在两个数据集上分别提升了1.4%和4.2%,且模型参数量仅为YOLOv5s的67.6%。

, correspAuthors=陈紫强, authorNote=null, correspAuthorsNote=
* 陈紫强(1973—),男,汉族,湖南沅江人,博士,副教授。研究方向:信号处理。E-mail:
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=EFysUFPLxwn+IzBZnOpUbQ==, magXml=ZI4UZyG62KOf0KMcoG4CUw==, pdfUrl=null, pdf=b8t7gRHHn5m3drzC4x27xA==, pdfFileSize=21855748, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=RSWGqMtqZrKQLAF5rJ30Pg==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=+3BnrBLOQxvn8nQxwvTniw==, mapNumber=null, authorCompany=null, fund=null, authors=

孔令鑫(1999—),男,汉族,江苏南京人,硕士研究生。研究方向:深度学习和图像处理。E-mail:

, authorsList=孔令鑫, 陈紫强, 晋良念, 蒋艳英)}, authors=[Author(id=1203787154905346520, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=1013653819@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1203787155052147177, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, authorId=1203787154905346520, language=EN, stringName=Ling-xin KONG, firstName=Ling-xin, middleName=null, lastName=KONG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China
2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1203787155236696575, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, authorId=1203787154905346520, language=CN, stringName=孔令鑫, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, address=1 桂林电子科技大学信息与通信学院, 桂林 541004
2 广西无线宽带通信与信号处理重点实验室, 桂林 541004, bio={"content":"

孔令鑫(1999—),男,汉族,江苏南京人,硕士研究生。研究方向:深度学习和图像处理。E-mail:

"}, bioImg=null, bioContent=

孔令鑫(1999—),男,汉族,江苏南京人,硕士研究生。研究方向:深度学习和图像处理。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1203787154393641372, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=1, ext=[AuthorCompanyExt(id=1203787154402029981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China), AuthorCompanyExt(id=1203787154410418592, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 桂林电子科技大学信息与通信学院, 桂林 541004)]), AuthorCompany(id=1203787154536247727, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=2, ext=[AuthorCompanyExt(id=1203787154544636337, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China), AuthorCompanyExt(id=1203787154553024947, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 广西无线宽带通信与信号处理重点实验室, 桂林 541004)])]), Author(id=1203787155375108624, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=1420536815@qq.com, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1203787155534492198, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, authorId=1203787155375108624, language=EN, stringName=Zi-qiang CHEN, firstName=Zi-qiang, middleName=null, lastName=CHEN, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, 3, *, address=1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China
2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China
3 Nanning Research Institute,Guilin University of Electronic Technology, Nanning 530000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1203787155656127025, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, authorId=1203787155375108624, language=CN, stringName=陈紫强, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, 3, *, address=1 桂林电子科技大学信息与通信学院, 桂林 541004
2 广西无线宽带通信与信号处理重点实验室, 桂林 541004
3 南宁桂电电子科技研究院有限公司, 南宁 530000, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1203787154393641372, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=1, ext=[AuthorCompanyExt(id=1203787154402029981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China), AuthorCompanyExt(id=1203787154410418592, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 桂林电子科技大学信息与通信学院, 桂林 541004)]), AuthorCompany(id=1203787154536247727, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=2, ext=[AuthorCompanyExt(id=1203787154544636337, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China), AuthorCompanyExt(id=1203787154553024947, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 广西无线宽带通信与信号处理重点实验室, 桂林 541004)]), AuthorCompany(id=1203787154750157252, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=3, ext=[AuthorCompanyExt(id=1203787154762740166, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154750157252, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 Nanning Research Institute,Guilin University of Electronic Technology, Nanning 530000, China), AuthorCompanyExt(id=1203787154771128776, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154750157252, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 南宁桂电电子科技研究院有限公司, 南宁 530000)])]), Author(id=1203787155756790332, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1203787155899396690, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, authorId=1203787155756790332, language=EN, stringName=Liang-nian JIN, firstName=Liang-nian, middleName=null, lastName=JIN, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, 3, address=1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China
2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China
3 Nanning Research Institute,Guilin University of Electronic Technology, Nanning 530000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1203787156000059999, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, authorId=1203787155756790332, language=CN, stringName=晋良念, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, 3, address=1 桂林电子科技大学信息与通信学院, 桂林 541004
2 广西无线宽带通信与信号处理重点实验室, 桂林 541004
3 南宁桂电电子科技研究院有限公司, 南宁 530000, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1203787154393641372, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=1, ext=[AuthorCompanyExt(id=1203787154402029981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China), AuthorCompanyExt(id=1203787154410418592, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 桂林电子科技大学信息与通信学院, 桂林 541004)]), AuthorCompany(id=1203787154536247727, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=2, ext=[AuthorCompanyExt(id=1203787154544636337, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China), AuthorCompanyExt(id=1203787154553024947, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 广西无线宽带通信与信号处理重点实验室, 桂林 541004)]), AuthorCompany(id=1203787154750157252, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=3, ext=[AuthorCompanyExt(id=1203787154762740166, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154750157252, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 Nanning Research Institute,Guilin University of Electronic Technology, Nanning 530000, China), AuthorCompanyExt(id=1203787154771128776, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154750157252, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 南宁桂电电子科技研究院有限公司, 南宁 530000)])]), Author(id=1203787156151054956, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1203787156285272694, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, authorId=1203787156151054956, language=EN, stringName=Yan-ying JIANG, firstName=Yan-ying, middleName=null, lastName=JIANG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, 3, address=1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China
2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China
3 Nanning Research Institute,Guilin University of Electronic Technology, Nanning 530000, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1203787156448850563, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, authorId=1203787156151054956, language=CN, stringName=蒋艳英, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, 2, 3, address=1 桂林电子科技大学信息与通信学院, 桂林 541004
2 广西无线宽带通信与信号处理重点实验室, 桂林 541004
3 南宁桂电电子科技研究院有限公司, 南宁 530000, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1203787154393641372, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=1, ext=[AuthorCompanyExt(id=1203787154402029981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China), AuthorCompanyExt(id=1203787154410418592, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 桂林电子科技大学信息与通信学院, 桂林 541004)]), AuthorCompany(id=1203787154536247727, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=2, ext=[AuthorCompanyExt(id=1203787154544636337, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China), AuthorCompanyExt(id=1203787154553024947, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 广西无线宽带通信与信号处理重点实验室, 桂林 541004)]), AuthorCompany(id=1203787154750157252, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=3, ext=[AuthorCompanyExt(id=1203787154762740166, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154750157252, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 Nanning Research Institute,Guilin University of Electronic Technology, Nanning 530000, China), AuthorCompanyExt(id=1203787154771128776, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154750157252, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 南宁桂电电子科技研究院有限公司, 南宁 530000)])])], keywords=[Keyword(id=1203787156742451866, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, orderNo=1, keyword=deep learning), Keyword(id=1203787156885058217, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, orderNo=2, keyword=object detection), Keyword(id=1203787157023470263, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, orderNo=3, keyword=road damages), Keyword(id=1203787157782639300, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, orderNo=4, keyword=YOLOv5s), Keyword(id=1203787157887496910, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, orderNo=5, keyword=deformable convolution), Keyword(id=1203787157996548824, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, orderNo=6, keyword=light weight), Keyword(id=1203787158130766568, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, orderNo=1, keyword=深度学习), Keyword(id=1203787158244012784, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, orderNo=2, keyword=目标检测), Keyword(id=1203787158327898873, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, orderNo=3, keyword=路面病害), Keyword(id=1203787158403396353, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, orderNo=4, keyword=YOLOv5s), Keyword(id=1203787158546002701, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, orderNo=5, keyword=可变形卷积), Keyword(id=1203787158680220445, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, orderNo=6, keyword=轻量化)], refs=[Reference(id=1203787165177196682, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=5220, pageEnd=5223, url=null, language=null, rfNumber=[1], rfOrder=0, authorNames=Wang W, Wu B, Yang S, journalName=IEEE International Conference on Big Data (Big Data), refType=null, unstructuredReference=Wang W, Wu B, Yang S, et al. Road damage detection and classification with faster R-CNN[C]// IEEE International Conference on Big Data (Big Data). Seattle: IEEE, 2018: 5220-5223., articleTitle=Road damage detection and classification with faster R-CNN, refAbstract=null), Reference(id=1203787165315608723, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2023, volume=24, issue=3, pageStart=3091, pageEnd=3103, url=null, language=null, rfNumber=[2], rfOrder=1, authorNames=Lin C, Tian D, Duan X, journalName=IEEE Transactions on Intelligent Transportation Systems, refType=null, unstructuredReference=Lin C, Tian D, Duan X, et al. DA-RDD: toward domain adaptive road damage detection across different countries[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(3): 3091-3103., articleTitle=DA-RDD: toward domain adaptive road damage detection across different countries, refAbstract=null), Reference(id=1203787165466603673, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=198, pageEnd=204, url=null, language=null, rfNumber=[3], rfOrder=2, authorNames=Ping P, Yang X, Gao Z, journalName=IEEE Sixth International Conference on Big Data Computing Service and Applications (BigDataService), refType=null, unstructuredReference=Ping P, Yang X, Gao Z, et al. A deep learning approach for street pothole detection[C]// IEEE Sixth International Conference on Big Data Computing Service and Applications (BigDataService). Oxford: IEEE, 2020: 198-204., articleTitle=A deep learning approach for street pothole detection, refAbstract=null), Reference(id=1203787165605015712, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2022, volume=23, issue=11, pageStart=22166, pageEnd=22178, url=null, language=null, rfNumber=[4], rfOrder=3, authorNames=Ma D, Fang H, Wang N, journalName=IEEE Transactions on Intelligent Transportation Systems, refType=null, unstructuredReference=Ma D, Fang H, Wang N, et al. Automatic detection and counting system for pavement cracks based on PCGAN and YOLO-MF[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11): 22166-22178., articleTitle=Automatic detection and counting system for pavement cracks based on PCGAN and YOLO-MF, refAbstract=null), Reference(id=1203787165726650533, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2021, volume=21, issue=15, pageStart=6380, pageEnd=6385, url=null, language=null, rfNumber=[5], rfOrder=4, authorNames=张世瑶, 贺玉彬, 周新志, journalName=科学技术与工程, refType=null, unstructuredReference=张世瑶, 贺玉彬, 周新志. 基于深度学习的道路表面裂缝检测方法[J]. 科学技术与工程, 2021, 21(15): 6380-6385., articleTitle=基于深度学习的道路表面裂缝检测方法, refAbstract=null), Reference(id=1203787165865062571, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2021, volume=21, issue=15, pageStart=6380, pageEnd=6385, url=null, language=null, rfNumber=[5], rfOrder=5, authorNames=Zhang Shiyao, He Yubin, Zhou, Xinzhi, journalName=Science Technology and Engineering, refType=null, unstructuredReference=Zhang Shiyao, He Yubin, Zhou, Xinzhi. Road surface crack detection method based on deep learning[J]. Science Technology and Engineering, 2021, 21(15): 6380-6385., articleTitle=Road surface crack detection method based on deep learning, refAbstract=null), Reference(id=1203787165982503088, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=1664, pageEnd=1671, url=null, language=null, rfNumber=[6], rfOrder=6, authorNames=Wang J, Gao X, Liu Z, journalName=IEEE 12th Data Driven Control and Learning Systems Conference (DDCLS), refType=null, unstructuredReference=Wang J, Gao X, Liu Z, et al. GSC-YOLOv5: an algorithm based on improved attention mechanism for road creak detection[C]// IEEE 12th Data Driven Control and Learning Systems Conference (DDCLS). Xiangtan: IEEE, 2023: 1664-1671., articleTitle=GSC-YOLOv5: an algorithm based on improved attention mechanism for road creak detection, refAbstract=null), Reference(id=1203787166078972084, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2023, volume=49, issue=2, pageStart=15, pageEnd=23, url=null, language=null, rfNumber=[7], rfOrder=7, authorNames=王朕, 李豪, 严冬梅, journalName=计算机工程, refType=null, unstructuredReference=王朕, 李豪, 严冬梅, 等. 基于改进YOLOv5的路面病害检测模型[J]. 计算机工程, 2023, 49(2): 15-23., articleTitle=基于改进YOLOv5的路面病害检测模型, refAbstract=null), Reference(id=1203787166200606904, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2023, volume=49, issue=2, pageStart=15, pageEnd=23, url=null, language=null, rfNumber=[7], rfOrder=8, authorNames=Wang Zhen, Li Hao, Yan Dongmei, journalName=Computer Engineering, refType=null, unstructuredReference=Wang Zhen, Li Hao, Yan Dongmei, et al. Pavement disease detection model based on improved YOLOv5[J]. Computer Engineering, 2023, 49(2): 15-23., articleTitle=Pavement disease detection model based on improved YOLOv5, refAbstract=null), Reference(id=1203787166318047423, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=1550, pageEnd=1554, url=null, language=null, rfNumber=[8], rfOrder=9, authorNames=Wei C, Zhang Q, Zhang X, journalName=IEEE International Conference on Big Data (BigData), refType=null, unstructuredReference=Wei C, Zhang Q, Zhang X, et al. YOLOv5s-BSS: a novel deep neural network for crack detection of road damage[C]// IEEE International Conference on Big Data (BigData). New York: IEEE Computer Society, 2023: 1550-1554., articleTitle=YOLOv5s-BSS: a novel deep neural network for crack detection of road damage, refAbstract=null), Reference(id=1203787166427099331, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2022, volume=22, issue=13, pageStart=5299, pageEnd=5305, url=null, language=null, rfNumber=[9], rfOrder=10, authorNames=罗晖, 余俊英, 涂所成, journalName=科学技术与工程, refType=null, unstructuredReference=罗晖, 余俊英, 涂所成. 基于深度学习的公路路面病害检测算法[J]. 科学技术与工程, 2022, 22(13): 5299-5305., articleTitle=基于深度学习的公路路面病害检测算法, refAbstract=null), Reference(id=1203787166548734152, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2022, volume=22, issue=13, pageStart=5299, pageEnd=5305, url=null, language=null, rfNumber=[9], rfOrder=11, authorNames=Luo Hui, journalName=Science Technology and Engineering, refType=null, unstructuredReference=Luo Hui, Yu Junying, Tu Suocheng. Pavement defect detection algorithm based on deep learning[J]. Science Technology and Engineering, 2022, 22(13): 5299-5305., articleTitle=Pavement defect detection algorithm based on deep learning, refAbstract=null), Reference(id=1203787166649397454, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2023, volume=59, issue=13, pageStart=305, pageEnd=316, url=null, language=null, rfNumber=[10], rfOrder=12, authorNames=倪昌双, 李林, 罗文婷, journalName=计算机工程与应用, refType=null, unstructuredReference=倪昌双, 李林, 罗文婷, 等. 改进 YOLOv7 的沥青路面病害检测[J]. 计算机工程与应用 2023, 59(13): 305-316., articleTitle=改进 YOLOv7 的沥青路面病害检测, refAbstract=null), Reference(id=1203787166750060760, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2023, volume=59, issue=13, pageStart=305, pageEnd=316, url=null, language=null, rfNumber=[10], rfOrder=13, authorNames=Ni Changshuang, Li Lin, Luo Wenting, journalName=Computer Engineering and Applications, refType=null, unstructuredReference=Ni Changshuang, Li Lin, Luo Wenting, et al. Disease detection of asphalt pavement based on improved YOLOv7[J]. Computer Engineering and Applications, 2023, 59(13): 305-316., articleTitle=Disease detection of asphalt pavement based on improved YOLOv7, refAbstract=null), Reference(id=1203787166842335448, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2022, volume=19, issue=7, pageStart=1927, pageEnd=1935, url=null, language=null, rfNumber=[11], rfOrder=14, authorNames=王保宪, 白少雄, 赵维刚, journalName=铁道科学与工程学报, refType=null, unstructuredReference=王保宪, 白少雄, 赵维刚. 基于特征增强学习的路面裂缝病害视觉检测方法[J]. 铁道科学与工程学报, 2022, 19(7): 1927-1935., articleTitle=基于特征增强学习的路面裂缝病害视觉检测方法, refAbstract=null), Reference(id=1203787166938804446, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2022, volume=19, issue=7, pageStart=1927, pageEnd=1935, url=null, language=null, rfNumber=[11], rfOrder=15, authorNames=Wang Baoxian, Bai Shaoxiong, Zhao Weigang, journalName=Journal of Railway Science and Engineering, refType=null, unstructuredReference=Wang Baoxian, Bai Shaoxiong, Zhao Weigang. Pavement crack damage visual detection method based on feature reinforcement learning[J]. Journal of Railway Science and Engineering, 2022, 19(7): 1927-1935., articleTitle=Pavement crack damage visual detection method based on feature reinforcement learning, refAbstract=null), Reference(id=1203787167052050661, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2020, volume=46, issue=null, pageStart=101182, pageEnd=null, url=null, language=null, rfNumber=[12], rfOrder=16, authorNames=Cao M T, Tran Q V, Nguyen N M, journalName=Advanced Engineering Informatics, refType=null, unstructuredReference=Cao M T, Tran Q V, Nguyen N M, et al. Survey on performance of deep learning models for detecting road damages using multiple dashcam image resources[J]. Advanced Engineering Informatics, 2020, 46: 101182., articleTitle=Survey on performance of deep learning models for detecting road damages using multiple dashcam image resources, refAbstract=null), Reference(id=1203787167152713959, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2021, volume=58, issue=14, pageStart=336, pageEnd=344, url=null, language=null, rfNumber=[13], rfOrder=17, authorNames=罗晖, 贾晨, 李健, journalName=激光与光电子学进展, refType=null, unstructuredReference=罗晖, 贾晨, 李健. 基于改进YOLOv4的公路路面病害检测算法[J]. 激光与光电子学进展, 2021, 58(14): 336-344., articleTitle=基于改进YOLOv4的公路路面病害检测算法, refAbstract=null), Reference(id=1203787167274348780, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2021, volume=58, issue=14, pageStart=336, pageEnd=344, url=null, language=null, rfNumber=[13], rfOrder=18, authorNames=Luo Hui, journalName=Advances in Laser and Optoelectronics, refType=null, unstructuredReference=Luo Hui, Jia Chen, Li Jian. Road surface disease detection algorithm based on improved YOLOv4[J]. Advances in Laser and Optoelectronics, 2021, 58(14): 336-344., articleTitle=Road surface disease detection algorithm based on improved YOLOv4, refAbstract=null), Reference(id=1203787168431976687, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2022, volume=23, issue=11, pageStart=22378, pageEnd=22385, url=null, language=null, rfNumber=[14], rfOrder=19, authorNames=Iparraguirre O, Iturbe-Olleta N, Brazalez A, journalName=IEEE Transactions on Intelligent Transportation Systems, refType=null, unstructuredReference=Iparraguirre O, Iturbe-Olleta N, Brazalez A, et al. Road marking damage detection based on deep learning for infrastructure evaluation in emerging autonomous driving[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23 (11): 22378-22385., articleTitle=Road marking damage detection based on deep learning for infrastructure evaluation in emerging autonomous driving, refAbstract=null), Reference(id=1203787168553611511, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, doi=null, pmid=null, pmcid=null, year=2021, volume=132, issue=null, pageStart=103935, pageEnd=null, url=null, language=null, rfNumber=[15], rfOrder=20, authorNames=Arya D, Maeda H, Ghosh S K, journalName=Automation in Construction, refType=null, unstructuredReference=Arya D, Maeda H, Ghosh S K, et al. Deep learning-based road damage detection and classification for multiple countries[J]. Automation in Construction, 2021, 132: 103935., articleTitle=Deep learning-based road damage detection and classification for multiple countries, refAbstract=null)], funds=[Fund(id=1203787164589994087, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, awardId=221035-K, language=CN, fundingSource=广西信息材料重点实验室基金(221035-K), fundOrder=null, country=null), Fund(id=1203787164808097907, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, awardId=20231011, language=CN, fundingSource=中央引导地方科技发展资金(20231011), fundOrder=null, country=null), Fund(id=1203787164950704249, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, awardId=2022KY0195, language=CN, fundingSource=广西高校中青年教师基础能力提升项目(2022KY0195), fundOrder=null, country=null)], companyList=[AuthorCompany(id=1203787154393641372, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=1, ext=[AuthorCompanyExt(id=1203787154402029981, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China), AuthorCompanyExt(id=1203787154410418592, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154393641372, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 桂林电子科技大学信息与通信学院, 桂林 541004)]), AuthorCompany(id=1203787154536247727, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=2, ext=[AuthorCompanyExt(id=1203787154544636337, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China), AuthorCompanyExt(id=1203787154553024947, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154536247727, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 广西无线宽带通信与信号处理重点实验室, 桂林 541004)]), AuthorCompany(id=1203787154750157252, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, xref=3, ext=[AuthorCompanyExt(id=1203787154762740166, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154750157252, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 Nanning Research Institute,Guilin University of Electronic Technology, Nanning 530000, China), AuthorCompanyExt(id=1203787154771128776, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, companyId=1203787154750157252, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 南宁桂电电子科技研究院有限公司, 南宁 530000)])], figs=[ArticleFig(id=1203787159133205313, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.1, caption=The overall structure of the LDC-YOLOv5 algorithm, figureFileSmall=+5jrIR8RtlE9zxrho1HKsA==, figureFileBig=GFJ7aoc3O0TDVhEg5fwIYQ==, tableContent=null), ArticleFig(id=1203787159275811657, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图1, caption=LDC-YOLOv5算法总体结构, figureFileSmall=+5jrIR8RtlE9zxrho1HKsA==, figureFileBig=GFJ7aoc3O0TDVhEg5fwIYQ==, tableContent=null), ArticleFig(id=1203787159401640787, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.2, caption=Schematic diagram of a deformable convolution, figureFileSmall=fyijB9DAzLoRttpe9ormFg==, figureFileBig=+YOBXd2mOdBb9Eu7LT97FA==, tableContent=null), ArticleFig(id=1203787159519081310, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图2, caption=可变形卷积示意图, figureFileSmall=fyijB9DAzLoRttpe9ormFg==, figureFileBig=+YOBXd2mOdBb9Eu7LT97FA==, tableContent=null), ArticleFig(id=1203787159636521834, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.3, caption=Comparison of conventional convolution and deformable convolution heat map, figureFileSmall=Cjklxo7rBB2aEa93r5zS0A==, figureFileBig=jYEBJJHyOALIvsyxVzj1yw==, tableContent=null), ArticleFig(id=1203787159795905401, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图3, caption=常规卷积与可变形卷积热力图对比图, figureFileSmall=Cjklxo7rBB2aEa93r5zS0A==, figureFileBig=jYEBJJHyOALIvsyxVzj1yw==, tableContent=null), ArticleFig(id=1203787159942706054, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.4, caption=Comparison of deep convolution and conventional convolutional structures, figureFileSmall=hZsBbdkoWCL/vyf6Uoe5zQ==, figureFileBig=8lUHl+SJMMl5PBkP1OUmvQ==, tableContent=null), ArticleFig(id=1203787160081118093, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图4, caption=深度卷积和常规卷积结构对比图, figureFileSmall=hZsBbdkoWCL/vyf6Uoe5zQ==, figureFileBig=8lUHl+SJMMl5PBkP1OUmvQ==, tableContent=null), ArticleFig(id=1203787160244695960, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.5, caption=DW_DCN module structure diagram, figureFileSmall=4DUjQentVNonF1D98oAJLg==, figureFileBig=BMVyXe4Fg07zpweYUgrg+w==, tableContent=null), ArticleFig(id=1203787160416662435, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图5, caption=DW_DCN模块结构图, figureFileSmall=4DUjQentVNonF1D98oAJLg==, figureFileBig=BMVyXe4Fg07zpweYUgrg+w==, tableContent=null), ArticleFig(id=1203787160609600429, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.6, caption=C3_Ghost module structure diagram, figureFileSmall=TgabYiIibQr126PNIUJEng==, figureFileBig=DrrQZrEM2HN8vbRgiYjB7g==, tableContent=null), ArticleFig(id=1203787160739623860, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图6, caption=C3_Ghost模块结构图, figureFileSmall=TgabYiIibQr126PNIUJEng==, figureFileBig=DrrQZrEM2HN8vbRgiYjB7g==, tableContent=null), ArticleFig(id=1203787160844481474, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.7, caption=TripletAttention module structure diagram, figureFileSmall=TG73o1bocpw/3lWrHmt7jw==, figureFileBig=o6U9DUcmiNdJ7Q8Z9YvjUA==, tableContent=null), ArticleFig(id=1203787160978699205, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图7, caption=TripletAttention模块结构图, figureFileSmall=TG73o1bocpw/3lWrHmt7jw==, figureFileBig=o6U9DUcmiNdJ7Q8Z9YvjUA==, tableContent=null), ArticleFig(id=1203787161112916942, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.8, caption=Training results with different final learning rates, figureFileSmall=h8E5w77bNOWB/F9oiDt9mg==, figureFileBig=cVqLNxriNfxkPrk3EoDNpA==, tableContent=null), ArticleFig(id=1203787161238746070, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图8, caption=不同最终学习率训练结果, figureFileSmall=h8E5w77bNOWB/F9oiDt9mg==, figureFileBig=cVqLNxriNfxkPrk3EoDNpA==, tableContent=null), ArticleFig(id=1203787161347797989, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.9, caption=Detection results of various algorithms on RDD2022 datasets, figureFileSmall=Qtx4RXgzJ1aN/i1XnYACwA==, figureFileBig=wDXYbC3QT93PBRTZ6ohXDw==, tableContent=null), ArticleFig(id=1203787161469432810, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图9, caption=各类算法在RDD2022数据集上的检测结果, figureFileSmall=Qtx4RXgzJ1aN/i1XnYACwA==, figureFileBig=wDXYbC3QT93PBRTZ6ohXDw==, tableContent=null), ArticleFig(id=1203787161586873327, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.10, caption=Detection results of various algorithms on Road Damage dataset, figureFileSmall=27TxoI8u2Qp0PTwxmjBZ4g==, figureFileBig=djXOck0mdeRvB1U+flTsMA==, tableContent=null), ArticleFig(id=1203787161725285371, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图10, caption=各类算法在Road Damage数据集上的检测结果, figureFileSmall=27TxoI8u2Qp0PTwxmjBZ4g==, figureFileBig=djXOck0mdeRvB1U+flTsMA==, tableContent=null), ArticleFig(id=1203787161846919168, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Fig.11, caption=Actual data test results, figureFileSmall=IaKOyfxZiCn6LFcaNuI8kw==, figureFileBig=64W1qXk/EIsp/2cgN3hZpw==, tableContent=null), ArticleFig(id=1203787162010497032, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=图11, caption=实际数据检测结果, figureFileSmall=IaKOyfxZiCn6LFcaNuI8kw==, figureFileBig=64W1qXk/EIsp/2cgN3hZpw==, tableContent=null), ArticleFig(id=1203787162165686290, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Table 1, caption=

Ablation experimental results of LDC-YOLOv5 algorithm on dataset RDD2022

, figureFileSmall=null, figureFileBig=null, tableContent=
DW_DCN C3_Ghost 三重注
意力
类别AP/% mAP
50/%
mAP
(50~95)/%
参数量/106 GFLOPs FPS
D00 D10 D20 D40
54.3 49.6 63.7 49.1 54.2 26.4 7.02 15.8 196
54.7 51.4 65.4 48.8 55.1 26.9 5.54 9.6 238
54.2 50.5 65.8 50.0 55.1 27.2 4.75 8.0 239
54.2 51.4 67.3 49.6 55.6 27.2 4.75 8.0 232
), ArticleFig(id=1203787162312486936, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=表1, caption=

LDC-YOLOv5算法在数据集RDD2022上的消融实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
DW_DCN C3_Ghost 三重注
意力
类别AP/% mAP
50/%
mAP
(50~95)/%
参数量/106 GFLOPs FPS
D00 D10 D20 D40
54.3 49.6 63.7 49.1 54.2 26.4 7.02 15.8 196
54.7 51.4 65.4 48.8 55.1 26.9 5.54 9.6 238
54.2 50.5 65.8 50.0 55.1 27.2 4.75 8.0 239
54.2 51.4 67.3 49.6 55.6 27.2 4.75 8.0 232
), ArticleFig(id=1203787162488647714, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Table 2, caption=

Compare the experimental results RDD2022 the algorithms on the dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 类别AP/% mAP
50/%
mAP
(50~95)/%
F1/% 参数量/106 GFLOPs FPS
D00 D10 D20 D40
SSD 44.7 37.6 77.7 39.6 49.7 24.0 41.7 14.8 25.0 103
YOLOv3-tiny 42.9 39.2 49.2 37.4 42.2 17.8 46.6 8.68 12.9 285
YOLOv5s 54.3 49.6 63.7 49.1 54.2 26.4 56.0 7.02 15.8 196
YOLOv7-tiny 51.3 43.6 63.7 43.6 50.5 23.3 52.7 6.23 13.9 200
PP-PicoDet-l 52.6 54.1 60.9 46.0 53.4 25.8 55.8 3.30 8.91 238
LDC-YOLOv5 54.2 51.4 67.3 49.6 55.6 27.2 57.2 4.75 8.0 232
), ArticleFig(id=1203787162627059753, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=表2, caption=

RDD2022数据集上的算法对比实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 类别AP/% mAP
50/%
mAP
(50~95)/%
F1/% 参数量/106 GFLOPs FPS
D00 D10 D20 D40
SSD 44.7 37.6 77.7 39.6 49.7 24.0 41.7 14.8 25.0 103
YOLOv3-tiny 42.9 39.2 49.2 37.4 42.2 17.8 46.6 8.68 12.9 285
YOLOv5s 54.3 49.6 63.7 49.1 54.2 26.4 56.0 7.02 15.8 196
YOLOv7-tiny 51.3 43.6 63.7 43.6 50.5 23.3 52.7 6.23 13.9 200
PP-PicoDet-l 52.6 54.1 60.9 46.0 53.4 25.8 55.8 3.30 8.91 238
LDC-YOLOv5 54.2 51.4 67.3 49.6 55.6 27.2 57.2 4.75 8.0 232
), ArticleFig(id=1203787162807414835, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Table 3, caption=

The experimental results of the algorithm on the Road Damage dataset are compared

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 类别AP/% mAP
50/%
mAP
(50~95)/%
F1/% 参数量/106 GFLOPs FPS
D00 D10 D20 D40
SSD 60.6 14.1 58.8 45.4 44.7 21.0 40.2 14.8 25.0 99
YOLOv3-tiny 53.5 42.4 69.1 65.0 57.5 23.1 60.4 8.68 12.9 285
YOLOv5s 60.2 49.1 65.4 73.9 62.1 28.9 63.0 7.02 15.8 178
YOLOv7-tiny 66.5 34.9 67.5 71.2 60.0 26.6 60.6 6.23 13.9 179
PP-PicoDet-l 59.3 42.7 51.2 75.2 57.2 23.0 60.0 3.30 8.91 223
LDC-YOLOv5 64.6 53.1 70.9 76.7 66.3 30.6 66.9 4.75 8.0 200
), ArticleFig(id=1203787164006985788, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=表3, caption=

Road Damage数据集上的算法对比实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 类别AP/% mAP
50/%
mAP
(50~95)/%
F1/% 参数量/106 GFLOPs FPS
D00 D10 D20 D40
SSD 60.6 14.1 58.8 45.4 44.7 21.0 40.2 14.8 25.0 99
YOLOv3-tiny 53.5 42.4 69.1 65.0 57.5 23.1 60.4 8.68 12.9 285
YOLOv5s 60.2 49.1 65.4 73.9 62.1 28.9 63.0 7.02 15.8 178
YOLOv7-tiny 66.5 34.9 67.5 71.2 60.0 26.6 60.6 6.23 13.9 179
PP-PicoDet-l 59.3 42.7 51.2 75.2 57.2 23.0 60.0 3.30 8.91 223
LDC-YOLOv5 64.6 53.1 70.9 76.7 66.3 30.6 66.9 4.75 8.0 200
), ArticleFig(id=1203787164141203524, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=EN, label=Table 4, caption=

LDC-YOLOv5 robustness verification results

, figureFileSmall=null, figureFileBig=null, tableContent=
标准差 mAP50/%
YOLOv5s LDC-YOLOv5
0.0 62.1 66.3
0.2 62.0 66.1
0.4 61.8 66.3
0.6 62.3 65.5
0.8 62.1 65.4
1.0 62.1 65.4
), ArticleFig(id=1203787164296392785, tenantId=1146029695717560320, journalId=1146123166801305609, articleId=1203753463323800184, language=CN, label=表4, caption=

LDC-YOLOv5鲁棒性验证结果

, figureFileSmall=null, figureFileBig=null, tableContent=
标准差 mAP50/%
YOLOv5s LDC-YOLOv5
0.0 62.1 66.3
0.2 62.0 66.1
0.4 61.8 66.3
0.6 62.3 65.5
0.8 62.1 65.4
1.0 62.1 65.4
)], attaches=null, journal=Journal(id=1146119176004939786, delFlag=0, nameCn=科学技术与工程, nameEn=Science Technology and Engineering, nameHistory1=null, nameHistory2=null, issn=1671-1815, eissn=, cn=11-4688/T, coden=null, periodic=4, language=CN, oaType=是, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=UKU/O7GSka5polgCTkbIIw==, journalPrice=null, startedYear=null, abbrevIsoEn=Sci Technol Eng, journalRemark=null, publicationField=null, createdTime=null, updatedTime=1754445529766, createdBy=null, updatedBy=13701087609, firstLetterCn=S, firstLetterEn=S, subjectCode=Natural Sciences, subjectName=自然科学, subjectCodeEn=Natural Sciences, subjectNameEn=null, picCn=UKU/O7GSka5polgCTkbIIw==, picEn=5hwlULoNwcbj3xUmVi9MAQ==, jcr=null, cjcr=null, exts=[JournalExt(id=1159791870395564357, language=CN, name=科学技术与工程, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=http://www.stae.com.cn/jsygc/home, createdTime=1754445529793, updatedTime=1754445529793, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=http://www.stae.com.cn/jsygc/site/menus/20090429150146001, submissionAuthorUrl=http://www.stae.com.cn/jsygc/author/login, submissionEditorUrl=http://www.stae.com.cn/jsygc/editor/login, submissionReviewUrl=http://www.stae.com.cn/jsygc/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1159791870441701702, language=EN, name=Science Technology and Engineering, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=null, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=http://www.stae.com.cn/jsygc/home, createdTime=1754445529804, updatedTime=1754445529804, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.stae.com.cn/jsygc/author/login, submissionEditorUrl=http://www.stae.com.cn/jsygc/editor/login, submissionReviewUrl=http://www.stae.com.cn/jsygc/reviewer/login, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1146123166801305609, websiteList=[Website(id=1148243202391400884, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146123166801305609, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/kxjsygc/CN, language=CN, createTime=1751692112777, createBy=18614031015, updateTime=1753520965431, updateBy=18614031015, name=科学技术与工程-中文站点, tplId=1146099689490845704, title=科学技术与工程, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1148622798802673703, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=articleTextType, value=kx, createTime=1751782615614, updateTime=1751782615614, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798781702180, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=banner, value=null, createTime=1751782615609, updateTime=1751782615609, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798769119267, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=j86gbwi+p0Idkyl5SzIlmQ==, createTime=1751782615606, updateTime=1751782615606, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798794285094, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1751782615612, updateTime=1751782615612, creator=18614031015, updator=18614031015), WebsiteProps(id=1148622798790090789, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1148243202391400884, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1751782615611, updateTime=1751782615611, creator=18614031015, updator=18614031015)]), Website(id=1155914124811976731, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1146123166801305609, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/kxjsygc/EN, language=EN, createTime=1753521003206, createBy=18614031015, updateTime=1753521003206, updateBy=18614031015, name=科学技术与工程-英文站点, tplId=1146101810881728533, title=Science Technology and Engineering, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1155914371227308235, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=articleTextType, value=kx, createTime=1753521061952, updateTime=1753521061952, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371210531016, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=banner, value=null, createTime=1753521061947, updateTime=1753521061947, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371202142407, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=logo, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic?fileId=j86gbwi+p0Idkyl5SzIlmQ==, createTime=1753521061945, updateTime=1753521061945, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371223113930, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/kjdb/CN/file/pic, createTime=1753521061950, updateTime=1753521061950, creator=18614031015, updator=18614031015), WebsiteProps(id=1155914371218919625, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1155914124811976731, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1753521061949, updateTime=1753521061949, creator=18614031015, updator=18614031015)])], journalTitle=科学技术与工程, weixinUrl=null, journalUrl=null, iacademicId=null, status=0, seqNo=null, journalTitleEn=Science Technology and Engineering, journalPhotoCn=UKU/O7GSka5polgCTkbIIw==, journalPhotoEn=5hwlULoNwcbj3xUmVi9MAQ==, journalFirstLetter=S, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=null, provinceCode=null, provinceName=null, collectFlag=false), detailUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2401676, detailUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/10.12404/j.issn.1671-1815.2401676, pdfUrlCn=https://castjournals.cast.org.cn/joweb/kxjsygc/CN/PDF/10.12404/j.issn.1671-1815.2401676, pdfUrlEn=https://castjournals.cast.org.cn/joweb/kxjsygc/EN/PDF/10.12404/j.issn.1671-1815.2401676, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
融合可变形卷积的轻量级路面病害检测算法
收藏切换
PDF下载
孔令鑫 1, 2 , 陈紫强 1, 2, 3, * , 晋良念 1, 2, 3 , 蒋艳英 1, 2, 3
科学技术与工程 | 论文·自动化技术、计算机技术 2025,25(2): 683-694
收起
收藏切换
科学技术与工程 | 论文·自动化技术、计算机技术 2025, 25(2): 683-694
融合可变形卷积的轻量级路面病害检测算法
全屏
孔令鑫1, 2 , 陈紫强1, 2, 3, * , 晋良念1, 2, 3, 蒋艳英1, 2, 3
作者信息
  • 1 桂林电子科技大学信息与通信学院, 桂林 541004
  • 2 广西无线宽带通信与信号处理重点实验室, 桂林 541004
  • 3 南宁桂电电子科技研究院有限公司, 南宁 530000
  • 孔令鑫(1999—),男,汉族,江苏南京人,硕士研究生。研究方向:深度学习和图像处理。E-mail:

通讯作者:

* 陈紫强(1973—),男,汉族,湖南沅江人,博士,副教授。研究方向:信号处理。E-mail:
A Lightweight Road Disease Detection Algorithm Fused with Deformable Convolution
Ling-xin KONG1, 2 , Zi-qiang CHEN1, 2, 3, * , Liang-nian JIN1, 2, 3, Yan-ying JIANG1, 2, 3
Affiliations
  • 1 School of Information and Communication, Guilin University of Electronic Technology, Guilin 541004, China
  • 2 Guangxi Key Laboratory of Wireless Wideband Communication and Signal Processing, Guilin 541004, China
  • 3 Nanning Research Institute,Guilin University of Electronic Technology, Nanning 530000, China
出版时间: 2025-01-18 doi: 10.12404/j.issn.1671-1815.2401676
文章导航
收藏切换

针对现有的路面病害检测算法在复杂环境下检测精度低、模型复杂度高的问题,在YOLOv5基础上,提出了一种融合可变形卷积的轻量级路面病害检测算法(lightweight deformable convolution YOLOv5,LDC-YOLOv5)。首先,针对真实路面病害复杂不规整的特点,使用可变形卷积(Deformable Conv)和深度卷积(Depthwise Conv),设计了一种轻量级特征提取模块,代替原网络主干部分的C3模块,使卷积核聚焦在无规则裂缝病害上,增强病害特征提取能力。其次,针对特征融合阶段出现算法复杂度过高的问题,使用轻量级卷积GhostConv,构建一种轻量级特征融合模块,代替原网络颈部网络部分的C3模块,降低网络参数和复杂度;为避免真实路面出现光照不均,出现阴影遮挡路面病害目标而造成的病害漏检的情况,在主干网络部分,引入轻量级注意力机制TripletAttention,增强算法对病害信息上下文之间的理解能力。最后在IEEE公开数据集RDD2022和Kaggle公开数据集Road Damage上进行测试,实验结果表明,与YOLOv5s相比,mAP50在两个数据集上分别提升了1.4%和4.2%,且模型参数量仅为YOLOv5s的67.6%。

深度学习  /  目标检测  /  路面病害  /  YOLOv5s  /  可变形卷积  /  轻量化

In response to the low detection accuracy and high model complexity of existing road damage detection algorithms in complex environments, a lightweight road damage detection algorithm named LDC-YOLOv5 (lightweight deformable convolution YOLOv5) was proposed based on YOLOv5.To address the complexity of real road surface damages, a lightweight feature extraction module was designed using Deformable Conv (deformable convolution) and Depthwise Conv (depthwise convolution) to replace the C3 module in the original network backbone, enabling convolutional kernels to focus on irregular crack damages and enhancing feature extraction for damage detection. To reduce algorithm complexity in the feature fusion stage, a lightweight feature fusion module was constructed using GhostConv to replace the C3 module in the original network neck, lowering network parameters and complexity. Additionally, to prevent missed detections caused by uneven lighting and shadow obstruction, a lightweight attention mechanism, TripletAttention, was introduced in the backbone network to improve the algorithm's understanding of damage information and context. Experiments conducted on the IEEE open dataset RDD2022 and the Kaggle open dataset Road Damage demonstrate that, compared to YOLOv5s, the proposed LDC-YOLOv5 achieves a 1.4% and 4.2% improvement in mAP50 on the two datasets, respectively, with only 67.6% of the model parameters of YOLOv5s.

deep learning  /  object detection  /  road damages  /  YOLOv5s  /  deformable convolution  /  light weight
孔令鑫, 陈紫强, 晋良念, 蒋艳英. 融合可变形卷积的轻量级路面病害检测算法. 科学技术与工程, 2025 , 25 (2) : 683 -694 . DOI: 10.12404/j.issn.1671-1815.2401676
Ling-xin KONG, Zi-qiang CHEN, Liang-nian JIN, Yan-ying JIANG. A Lightweight Road Disease Detection Algorithm Fused with Deformable Convolution[J]. Science Technology and Engineering, 2025 , 25 (2) : 683 -694 . DOI: 10.12404/j.issn.1671-1815.2401676
随着中国经济的增长,路面的通车量持续增加,路面显性病害日益增多。传统的路面病害检测方式存在耗时且劳动密集的问题,难以适用于当前车流量高的道路,因此如何便捷高效地实现路面病害检测不仅仅是当前中外热点研究问题,而且是保障日常交通出行安全的重要需求。
近年来,随着人工智能技术的快速发展,基于深度学习的路面病害检测逐渐引人关注,中外学者在此领域展开广泛深入研究。当前的深度学习目标检测算法可分为双阶段检测算法和单阶段检测算法两大类,双阶段算法检测有两个阶段:第一阶段是选择产生可能的物体候选区域;第二个阶段是对所产生的候选区域进行分类识别。Wang等[1]对数据样本进行分析,增强训练使用数据情况,调整候选框大小,最后在Faster R-CNN上取得较高的检测结果。Lin等[2]将Faster R-CNN与图像级和实例级对齐方法相结合,以对抗方式用于领域自适应路面伤损检测,提升检测精度。这类算法检测精度较高,但是模型庞大,不适合用在实时处理系统中。单阶段检测的算法精度相对低于双阶段算法,但是算法参数量较小,复杂度较低,适合应用于轻量化路面病害检测。Ping等[3]使用SSD算法进行路面病害检测,实验表明SSD算法与YOLO系列算法检测精度相当,但网络参数高于YOLO系列算法。Ma等[4]在YOLOv3主干网络使用5个特征提取模块,同时删除最大池化层并添加更多1×1卷积层,虽然获得更好的病害特征表达以及较快的检测速度,但是网络参数依旧较高。张世瑶等[5]提出一种针对道路表面像素级检测方法,使用卷积神经网络对裂缝原始图像进行分割、二值化等一系列处理,最后在U-Net++网络上进行验证,得到了较高的检测准确率。Wang等[6]通过在YOLOv5的快速金字塔池化模块SPPF(spatial pyramid pooling-fast,SPPF)前面添加改进洗牌注意力机制模块IMSA(Improved Shuffle Attention,IMSA),并添加了残差块实现特征融合,在不增加网络参数量的情况下提高检测精度。王朕等[7]在YOLOv5模型中引入坐标注意力机制增强模型对感兴趣的区域特征提取能力,同时将自注意力机制整合到C3模块中,提取更多差异化特征,实现检测能力的增强。Wei等[8]使用双向特征金字塔网络和空间金字塔池跨阶段部分连接模块SPPCSPC(spatial pyramid pooling cross-stage partial connections,SPPCSPC)代替YOLOv5网络原有的特征连接方式,研究结果表示,修改后的网络在不降低检测速度的情况下,检测精度略微提升。以上算法在提高路面病害检测精度方面取得了良好的进展,然而在实际工程应用中,除了考虑精度外,还必须兼顾模型复杂度和实时性的要求,使算法适用于小型边缘计算设备。
真实的路面裂缝病害具有走向随机、形状复杂且不固定的特点。上述文献算法在特征提取时,使用常规卷积不能有效地提取真实路面裂缝病害目标特征信息;其次在对范围较大目标进行特征提取和融合时,上述文献使用大量常规卷积会产生较高的参数量和复杂度,不适合在路面病害检测中实现轻量化,无法达到在小型平台上进行实时检测的目的。
YOLOv5目标检测算法是YOLO系列相对成熟的经典算法之一,和其他系列目标检测算法[9-14]相比,YOLOv5算法在检测精度和检测速度之间都有着较好的均衡,同时骨干网络采用CSPNet结构有助于减轻梯度消失问题,提高信息传递的效率。并且YOLOv5算法根据不同检测任务,有着YOLOv5s等5个结构相同,参数量与复杂度不同的模型,更加方便模型在不同平台上的移植。鉴于YOLOv5在目标检测领域存在巨大优势,现提出一种轻量级路面病害检测算法LDC-YOLOv5。针对YOLOv5算法存在对路面病害目标特征提取能力不足的问题,以及模型参数量和复杂度仍然较高,不利于模型在小型嵌入式平台上移植的问题,做出以下改进:①使用可变形卷积和深度卷积构建一种轻量级特征提取模块,对卷积核增添可学习的偏移量,使得卷积核能够聚焦在形变无规则路面裂缝病害上,克服常规卷积对非规整目标特征提取不足的缺点;②使用轻量级卷积GhostConv构成一种轻量级特征融合模块,将原来特征融合模块在通道数较高位置的常规卷积替换为少量卷积和深度卷积的叠加,有效缓解特征融合阶段参数和复杂度过大的问题;③面对真实路面会出现光照不均,可能会出现阴影遮挡路面病害目标而造成的病害漏检的情况,引入轻量级的注意力机制TripletAttention,增强上下文理解,进一步增强网络特征提取能力。
虽然YOLOv5以模型轻,检测精度高,在目标检测领域有较大优势,但是对于路面病害形变复杂,环境影响严重的情况下,难以达到轻量化且高精度的效果。因此本文研究提出一种融合可变形卷积的轻量级路面病害检测算法LDC-YOLOv5,算法结构如图1所示。
以YOLOv5为基础,在主干网络提出一种轻量级特征提取模块DW_DCN,不仅使用深度卷积降低网络参数,而且使用可变形卷积提高卷积对形变复杂的路面病害提取能力。在颈部网络,设计一种轻量级特征融合模块C3_Ghost,通过GhostConv减少网络融合中产生的庞大参数和复杂度。为了降低光照不均影响病害漏检缺陷率,最后在主干网络引用轻量级注意力机制模块TripletAttention,增强上下文理解能力。
真实路面裂缝病害充满随机性,如果使用常规卷积,大多数卷积核只会卷积在背景区域,无法获得路面病害目标的主要特征信息。可变形卷积则是在卷积核中添加了可学习的偏移量,在卷积训练时,卷积核可以始终与物体的实际形状贴近。
在常规卷积中,输出特征的表达式为
y(p0)= p n Rw(pn)x(p0+pn)
式(1)中:w(·)为卷积核所对应的权重值;x(·)为对应卷积核位置的特征值;p0为卷积中心点;pnp0在卷积和范围内每个偏移量的位置,在可变形卷积中引入可学习的Δpn偏移量,并且偏移赋予λpn权重,这样偏移后,卷积核可以更好地贴合目标,从而获得更多有效的特征信息。因此输出特征表达式为
y(p0)= p n Rw(pn)x(p0+pn+Δpn)λpn
由于加入的偏移量一般都是小数,使得输入项x(·)中的表达式为非整数,即表达的并不是图像上对应的像素点,因此需要使用双线性插值获得偏移后的像素点,其表达式为
$\begin{aligned} x(p) & =\sum_{q=0}^{1} G(q, p) x(q) \\ & =\sum_{q=0}^{1} g\left(q_{x}, p_{x}\right) g\left(q_{y}, p_{y}\right) x(q) \\ & =\sum_{q=0}^{1} \max \left(0,1-\left|q_{x}-p_{x}\right|\right) \max (0,1- \\ & \left.\left|q_{y}-p_{y}\right|\right) x(q) \end{aligned}$
式(3)中:G(·)为双线性插值核;pq分别为空间中任意位置和枚举特征映射x中所有积分空间位置。
通常将插值点位置的像素值设置为其领域4个像素点的加权和,每一个领域点权重根据自身与插值点横纵坐标距离来设置,使用max表达式即限制了插值点与领域点不会超过一个像素的距离。
运用可变形卷积可以在面对形变复杂的路面裂缝病害中,充分提取型变复杂的特征信息,学习到每个卷积块的偏移量和权重,如图2所示。
真实路面裂缝病害走向随机,形状复杂,在大多数检测算法中,使用常规卷积难以有效提取其特征,如图3常规卷积热力图和可变性卷积热力图对比所示,常规卷积在走向较为规整,特征较为明显的裂缝处,能够提取有效特征信息,而形状复杂的龟裂病害,常规卷积难以有效提取其特征。但是可变性卷积则可以避免这个缺陷。
考虑到使用可变形卷积会添加了额外的学习量,网络的参数和复杂度得到提升,不利于网络轻量化处理,因此在使用可变形卷积前,使用深度卷积将通道数进行缩放。
与常规卷积不同,深度卷积的每一个卷积核只负责一个通道卷积,因此每个卷积核所需参与计算量大幅度降低。将深度卷积生成的若干特征图进行顺序拼接便得到与输入通道相同的特征图,深度卷积和常规卷积结构对比图如图4所示。
轻量级特征提取模块DW_DCN如图5所示,首先通过一个是3×3的DWConv将通道数扩张,接着通过一个1×1的常规卷积压缩通道数,并送入由深度卷积和可变形卷积构成的Bottleneck_DCN模块,在Bottleneck_DCN模块中,先通过一个1×1的DWConv进行通道压缩,再进行一个3×3的可变形卷积提取形变复杂的病害特征,之后通过一个1×1的DWConv进行通道复原,最后通过与初始输入相加构成残差块。通过Bottleneck_DCN模块后,与最先由DWConv输出后一个降低通道的1×1常规卷积进行Concat操作,使得通道数叠加。最后在进行一个1×1的常规卷积将输出通道数转换为所需要求。DW_DCN模块在保证获得丰富的信息流的同时,使用更为轻量化的DWConv降低参数量,达到轻量化特征提取的目的。
轻量级卷积GhostConv相对于常规卷积所产生的参数量更少,复杂度更低。GhostConv使用少量卷积和线性运算进行叠加而得到输出特征。
假设输入特征的维度为c×h×w,使用常规卷积,经过nk×k的卷积核处理后输出特征维度为n×h'×w',其卷积参数量和运算复杂度为
P=nckk
C=nh'w'ckk
式中:P为常规卷积参数量;C为常规卷积参数量。常规卷积层的参数量只与卷积核的数量、输入通道数和卷积核的大小有关,与特征图的空间出度无关,因此可用n×c×k×k表示;n×h'×w'为输出特征图的总元素数量;c×k×k为计算一个输出位置是所需的乘加操作数,因此常规卷积计算量可有两者相乘所得。
同样的输入与输出,在Ghost卷积中,则会先使用一个少量卷积核进行常规卷积,得到(n/sh'×w'本征特征输出图;接着对本征特征输出图进行s-1次线性变换,生成(s-1)×(n/sh'×w'的Ghost特征图;最后将本征特征输出图与Ghost特征图进行拼接得到相同的输出,其卷积参数量和运算复杂度为
PG= n sckk+(s-1) n sdd
CG= n sh'w'ckk+(s-1) n sh'w'dd
式中:PG为Ghost卷积参数量;CG为Ghost卷积计算量;s为总线性变换次数,s远小于通道数c;n/s为第一次变换时输出通道;d×d为线性运算的平均核大小,d×d大小远小于k×k
两种卷积的计算量和复杂度进行对比可知
rP= P G P= n s c k k + ( s - 1 ) n s d d n c k k s + c - 1 s c 1 s
rC= C G C= n s h ' w ' c k k + ( s - 1 ) n s h ' w ' d d n h ' w ' c k k s + c - 1 s c 1 s
式中:rP为Ghost卷积与常规卷积参数量对比值;rC为Ghost卷积与常规卷积计算量对比值。
相比于常规卷积,线性运算的成本可以忽略不计。因此在得到相同大小的特征图时,使用Ghost卷积能够大量减少计算参数和复杂度。
轻量级特征提取模块C3_Ghost在原来C3基础上,使用由Ghost卷积构成轻量级的Bottleneck_Ghost模块替换原先的Bottleneck模块。Bottleneck_Ghost模块由两个Ghost卷积组成,第一个用于通道扩张,获得更多特征信息,另一个用于压缩通道,最后与初始输入构成残差连接。C3_Ghost模块如图6所示,首先通过一个1×1的卷积层是输入通道数压缩一半,再输送至Bottleneck_Ghost模块进行卷积运算,之后将C3_Ghost的输入进行额外的一个1×1的卷积层与Bottleneck_Ghost模块的输出进行Concat操作,最后由一个1×1的卷积层改变输出通道数得到输出。
由于路面病害检测对象是真实的路面环境,不可避免地会出现光照不均而引发的路面病害被漏检误检的现象,干扰网络对病害的检测能力。因此,在设计的主干网络中最后一个DW_DCN模块后,添加一个轻量级注意力机制模块TripletAttention,其使用三重分支结构捕捉不同维度交互获得注意权重,增强网络的感受野和上下文理解能力。同时TripletAttention强调了空间维度之间的联系,弥补了特征提取时使用DWConv而忽视通道之间的联系的不足。
TripletAttention的3个分支,其中两个分支用于捕获通道和空间W以及H维度之间的交互关系,另一个分支用于处理注意力权重的计算。其结构如图7所示。
第一个分支用于通道数C和空间H维度的交互,首先进行维度转换,接着进行Z-pool操作,通过最大池化层和平均池化层双重结合使得空间维度H变换为2,再通过一个7×7的卷积将空间维度变成1,之后通过Sigmoid函数生成该分支的注意力权重,最后将维度转换为初始状态。第二个分支和第一个类似,是关于通道数C和空间W维度的交互。第三个分支是空间注意力权重的计算,不需要通过维度转换。最后是计算3个分支的平均值。
在原网络YOLOv5s中,主干网络采用了CSPDarknet53经典结构,包含4个C3特征提取模块。C3模块包含大量常规卷积,随着输入特征通道数增加,该模块产生的参数量和计算量会急剧增大。同样,在颈部网络部分中,YOLOv5s使用PANet网络结构,包含4个C3特征融合模块,参数量和复杂度也非常大。
LDC-YOLOv5算法以YOLOv5s算法为基础,在主干网络使用轻量级深度卷积替换C3模块中用于通道升高的常规卷积操作。同样,在颈部网络,利用Ghost卷积替换C3模块Bottleneck部分用于通道变化的常规卷积项。两种卷积的替换,使得LDC-YOLOv5算法在原网络检测精度保持不变情况下,大幅减少了算法的复杂度。具体参数量(params)和浮点计算量(GFLOPs)的定量分析如表1所示。
RDD2022数据集是IEEE公开挑战赛所提供的路面病害数据集,该数据集包含了由日本、印度、捷克共和国、挪威、美国和中国6个国家,以及晴天、阴天、夜晚等多场景共47 420幅使用智能手机拍摄的路面图像,其中主要涉及了纵向裂缝(D00)、横向裂缝(D10)、龟裂(D20)和坑洞(D40)这4种主要路面病害。为了方便实验,使用了RDD2022数据集中由官方已经标注好xml标签的38 385张数据训练集作为实验时使用的数据集,按照8∶1∶1的比例划分为训练集、验证集和测试集。这种划分方法可以保证训练集拥有足够的数据样本用于训练,同时保留了相等数量的验证和测试数据,使得模型性能评估更具可靠性。该方法也是目前深度学习算法中最常见的数据划分方式之一。
Road Damage数据集是Kaggle平台提供的公开路面病害数据集,收集者Alvaro Basily使用小米手机拍摄高分辨率印度尼西亚路面病害,也包含晴天、阴天等多场景共计3 321张图片数据,同样也涉及了D00、D10、D20、D40 4种路面病害。按照8∶1∶1的比例划分为训练集、验证集和测试集,用于进一步验证本文算法对路面病害检测的强大能力,以确保实验的科学性和模型评估的可靠性。
该数据集是由广东省政务服务数据管理局,于2021年举办2021全球开放数据应用创新大赛中,道路路面病害智能分析算法赛道所提供。该数据集包含14 000张车载摄像头拍摄实际的中国城市道路病害图像样本。使用其中测试集A作为实际路面病害检测案例数据,用于分析本文LDC-YOLOv5算法训练模型的有效性。
本文实验环境:操作系统为 Ubuntu20.04,GPU为NVIDIA RTX 3060,内存为64 G,Pytorch 版本为 2.0.0,cudnn为11.7。在训练时,采用 SGD 优化器对网络进行优化,同时使用Mosaic数据增强方法,随机将训练时使用4张图片进行拼接,并保留拼接图片中的目标信息,增强模型的分类表现和鲁棒性。实验采用单个类别的平均精度(average precision,AP)、多个类别的平均精度(mean average precision,mAP)、F1分数、模型参数(parmas)和浮点运算次数(GFLOPs)对模型进行评估,IoU 阈值为0.7,每个模型训练300轮。为了统一变量,将输入图像尺寸设为640×640进行算法比较。
初始学习率的范围一般在0.1~0.001,其目的是让训练时的损失尽可能快地降低到最佳范围内。目前公认的初始学习率使用值为0.01,该数值可以有效避免模型在训练初期出现较大的波动和收敛过慢导致训练所需时间增长现象。因此本次实验初始学习率设定为0.01。
为了选择最佳的最终学习率,对最终学习率进行敏感性训练分析。参考YOLOv5官方文档所给出的最终学习率为0.000 1和0.001两种选择方案,同时参考全球道路损坏检测挑战赛2020冠军队伍在YOLOv5模型上所使用最终学习率0.002方案。使用本文算法在Road Damage数据集,分别使用最终学习率为0.000 1、0.000 5、0.001、0.001 5、0.002的方案,每个方案进行了300轮训练。以每轮训练出的模型在验证集上的mAP50作为参考指标,选择训练效果最佳的最终学习率作为模型训练的最佳学习率指标,其结果如图8所示。
通过验证结果可知,随着学习轮数增加,不同学习率下的模型验证指标mAP50都会向着一个极大值靠近,当选择最终学习率为0.001时,验证指标mAP50可以达到最大,说明LDC-YOLOv5算法选择最终学习率为0.001时,能够生成性能最佳的模型,因此选择初始学习率为0.01,最终学习率为0.001作为实验训练指标。
为了验证LDC-YOLOv5算法各个模块的有效性,在数据集RDD2022上进行消融实验,实验结果如表1所示。使用DW_DCN模块后,网络整体参数下降约21.0%,浮点运算次数即复杂度减少39.3%左右,平均精度mAP50上升0.9%,mAP50-95上升0.5%,FPS上升42帧较为明显。在这模块中使用可变形卷积,对裂缝类病害形变复杂无规律的目标特征提取能力增强,3种裂缝类的病害检测精度提高,尤其是对占地较大的横向裂缝和龟裂块检测精度提升较大。可变形卷积的采样位置更符合目标本身的形状好和尺寸,而常规卷积难以做到。使用C3_Ghost模块后,网络参数再次下降11.2%左右,复杂度下降11.1%左右,平均精度mAP50并没有太大变化,mAP50-95上升0.3%,FPS几乎不变,该模块在特征融合阶段降低了特征冗余问题,保留主要特征信息,横向裂缝和纵向裂缝特征信息略微减少,检测精度降低,但是总体相对稳定。在引入注意力机制后,网络模型的参数和复杂度并没有发生增长,而检测平均精度mAP50上升0.5%,mAP50-95没有变化,注意力机制弥补了先前模块中使用深度卷积而造成的空间维度之间相关性信息丢失的缺陷,对与横向裂缝与龟裂病害上下文信息差异较大的目标提升较大,检测精度得到进一步提升。但是引用注意力机制会增加网络的深度,使得检测速度FPS(frames per second,FPS)略微降低。
本次实验在数据集RDD2022对LDC-YOLOv5算法和其他经典轻量级目标检测算法进行对比,验证本文算法的有效性,结果如表2所示。
表2结果可看出,从检测精度来分析,本文算法LDC-YOLOv5的平均检测精度mAP50达到55.6%,超过其他对比算法。YOLOv5s平均精度mAP50次之,达到54.2%,其次分别是PP-PicoDet-l、YOLOv7-tiny、SSD和YOLOv3-tiny,平均精度mAP50分别为53.4%、50.5%、49.7%和42.2%。4种路面病害中,横向裂缝、纵向裂缝和龟裂这3种裂缝类病害走势细长,形状不规则,特征较难表达。其他经典算法在特征提取阶段仅使用常规卷积不能得到较高的检测精度,本文算法使用可变形卷积能够有效提取裂缝类的形变复杂的病害特征,从而在实验结果中,总体检测精度与其他算法相比较高。但是裂缝类病害特征相似,除了龟裂病害会出现范围较大的目标体外,纵向裂缝和横向裂缝仅走势不同,较难区别。因此本文算法虽然可以提高对裂缝的检测率,但是区别细致的裂缝病害仍有瑕疵,所以对于纵向裂缝的检测精度,本文研究在与基础网络YOLOv5s的对比上,并没有较高提升。对于坑洞类较为形状常规病害目标,由于本文算法的网络层数略微增加,使得该病害检测精度有小幅度上升;从网络参数来分析,本文算法LDC-YOLOv5网络参数仅有4.75 M,分别是YOLOv7-tiny、YOLOv5s、YOLOv3-tiny和SSD算法的76.2%、67.7%、54.7%和32.1%,而PP-PicoDet-l算法网络结构较轻,参数量仅有3.3 M,远超其他算法;从推理速度进行分析,所有的算法在本实验环境下,FPS值均超过30,可达到实时检测需求,YOLOv3-tiny网络凭借其网络层数最少的优点,FPS能达到285,PP-PicoDet-l算法次之,可达238。本文算法略逊于PP-PicoDet-l算法,FPS为232,YOLOv7-tiny和YOLOv5s相近,均在200左右,SSD算法较差,仅有103。综上分析,本文算法在路面检测需求中占据较高的优势。
图9是这些算法在RDD2022数据集上检测结果。从3张对比图中,不难发现其他经典目标检测算法对路面病害都存在一定的漏检情况,在第一张图中,由于远处的坑洞受拍摄角度的影响,其边缘特征不规律且不够明显。为了充分提取目标特征信息,本文算法采用了可变形卷积,将卷积核聚焦在病害目标周围。相比于其他算法,使用常规卷积可能导致各种漏检情况,而本文算法可以检测出所有的坑洞病害。在第2张对比图中,横向裂缝范围较小,且在沥青路面上颜色特征不明显,导致许多经典目标检测算法出现漏检情况。然而,本文算法可以检测出所有的病害。在第3张对比图中,树荫下的裂缝目标受到光线照射不均的影响,其特征信息与树枝阴影难以区分。常规卷积难以有效提取出该目标的特征信息,因此其他算法均未能检测出此目标。然而本文算法结合了可变形卷积与注意力机制,能够有效区分裂缝与阴影之间的差别,并成功提取出在阴影中形变复杂的裂缝目标特征,避免了漏检现象的出现。
为验证LDC-YOLOv5算法的泛化能力以及在实际工程运用中的有效性,本文研究在高分辨路面病害数据集Road Damage上进行对比试验,实验结果如表3所示。
表3可发现,本文算法LDC-YOLOv5对于高分辨率的路面病害检测情况依旧领先,本文算法在Road Damage上mAP50可达66.3%。YOLOv5s算法次之,mAP50为62.1%,其次分别YOLOv7-tiny算法、YOLOv3-tiny算法、PP-PicoDet-l算法和SSD算法,mAP50分别为60.0%、57.5%、57.2%、44.7%。在单个类别病害比较中,本文算法相较于原YOLOv5s网络,对所有病害类别的检测精度均有显著提升。这得益于可变形卷积在对复杂裂缝类病害进行特征提取时的卓越性能。具体而言,裂缝类3种病害的检测精度均增长超过4%,而坑洞病害目标的提升略低于裂缝类。在纵向裂缝方面,YOLOv7-tiny算法表现出最高的检测精度,而YOLOv3-tiny算法最低,这表明算法网络层数的增加对病害检测有一定的影响;横向裂缝和龟裂病害的检测中,SSD算法和PP-PicoDet-l算法检测精度最低,而本文算法则取得最高的精度。这表明对于形变较为复杂的裂缝类病害,应用常规卷积进行病害特征提取可能无法获得令人满意的效果,而本文算法利用可变形卷积的优势再次彰显;坑洞类病害在实际场景中形状也是各异的,本文算法在检测上也具备一定优势,因此检测精度最高。实验结果表明,LDC-YOLOv5算法在不同数据集上展现出较强的泛化能力,病害检测精度相对较高,在实际工程运用中占据绝对优势。
图10分别是这些算法在Road Damage数据集上检测结果。本文算法可以降低对病害漏检或者误检情况,特别是在第三张对比图中,本文算法并没有将车道中间的水渍造成色彩影响部分检测成病害目标,同时能够正确检测出白色面包车附近的裂缝病害目标,再次证明了本文算法对路面病害检测可靠有效性。但是本文算法在对小目标的检测上还有一定的劣势,在第一张对比图中,本文算法不能检测出靠近汽车附近较小的坑洞病害。
为了验证LDC-YOLOv5算法的鲁棒性能,对Road Damage数据集的测试集添加标准差不同,均值为0的高斯噪声,用于评估本文模型在对抗样本上的表现能力。为了控制噪声仅产生微小的像素级扰动,不会对模型的输出产生显著影响,控制高斯噪声的标准差在0.2、0.4、0.6、0.8和1.0时,对模型输入数据产生干扰。同时将鲁棒性较好的YOLOv5s模型作为比较,以map50为参考结果,从而判断LDC-YOLOv5算法模型的鲁棒性能,其结果如表4所示。
表4可知,在受到不同强度的高斯噪声影响时,LDC-YOLOv5算法依然可以取得较高mAP50数值,其均值为65.74%。而YOLOv5s模型在受到噪声时获得的均值为62.06%,本文模型均值比YOLOv5s模型高3.68%。从稳定性上分析,YOLOv5s模型面对不同噪声时mAP50数值的方差为0.182,而本文模型为0.428。在高斯噪声的标准差为0.6时,本文模型检测结果略微下降。总的来说,与鲁棒性较强的YOLOv5s模型相比,本文模型在面对不同噪声影响时,依旧可以保持较高的检测精度,但是随着噪声强度的加强,本文算法模型会比YOLOv5s模型出现波动较大,但是检测精度仍比YOLOv5s模型高,总体性能较好,鲁棒性较强。
为了进一步验证LDC-YOLOv5算法模型的有效性,使用2021全球开放数据创新应用大赛 “道路路面病害智能分析算法”数据集中测试集A作为实际的路面病害检测案例,对LDC-YOLOv5算法在RDD2022数据集上训练完成的模型进行分析验证。由于该数据集中病害标签与RDD2022数据病害标签相差较大,模型无法直接计算出mAP50等相关指标,因此添加YOLOv5s算法在RDD2022数据集上训练完成的模型进行对比参照,其结果如图11所示。
图11可看出,LDC-YOLOv5模型在实际场景下具有良好的检测性能,模型的鲁棒性和实用性较高。在第一张和第三张对比图中,YOLOv5s算法模型会出现一定程度的漏检情况,甚至在第二张图中出现将栅栏影子误检成龟裂病害的情况。但是本文模型均不会出现这种错误,整体性能较强。
聚焦于路面病害检测模型检测精度低且模型复杂的问题,提出了一种融合可变形卷的轻量级路面检测病害算法LDC-YOLOv5。通过实验对所提出的算法进行验证,得到以下结论。
(1)为解决常规卷积对形变复杂的裂缝类病害特征提取能力不足的问题,本文算法在主干网络部分使用可变形卷积和深度卷积,构建轻量级特征提取模块,可有效提取形变复杂、走向随机的裂缝类病害特征同时,降低网络参数。
(2)为缓解特征融合部分模型复杂度过高的问题。使用轻量级卷积Ghost卷积构建轻量级特征提取模块,在不降低检测精度的同时,可以有效减少模型整体参数。
(3)实验结果表明,与其他经典轻量级目标检测算法相比,所提出的算法相比于原算法YOLOv5s在数据集RDD2022和Road Damage上mAP50指标分别提升1.4%和4.2%,且模型参数仅为原网络的67.6%。所提出算法更易于部署在边缘计算平台上实现实时病害检测任务。虽然本文模型在检测道路病害上有着极大有事,但面对噪声的侵扰会场生一定波动。未来将改进算法,增强算法对数据异常的处理能力,同时收集大量真实路面病害数据用于模型的训练,进一步提升算法鲁棒性和泛化能力。
  • 广西信息材料重点实验室基金(221035-K)
  • 中央引导地方科技发展资金(20231011)
  • 广西高校中青年教师基础能力提升项目(2022KY0195)
参考文献 引证文献
排序方式:
[1]
Wang W, Wu B, Yang S, et al. Road damage detection and classification with faster R-CNN[C]// IEEE International Conference on Big Data (Big Data). Seattle: IEEE, 2018: 5220-5223.
[2]
Lin C, Tian D, Duan X, et al. DA-RDD: toward domain adaptive road damage detection across different countries[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(3): 3091-3103.
[3]
Ping P, Yang X, Gao Z, et al. A deep learning approach for street pothole detection[C]// IEEE Sixth International Conference on Big Data Computing Service and Applications (BigDataService). Oxford: IEEE, 2020: 198-204.
[4]
Ma D, Fang H, Wang N, et al. Automatic detection and counting system for pavement cracks based on PCGAN and YOLO-MF[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11): 22166-22178.
[5]
张世瑶, 贺玉彬, 周新志. 基于深度学习的道路表面裂缝检测方法[J]. 科学技术与工程, 2021, 21(15): 6380-6385.
Zhang Shiyao, He Yubin, Zhou, Xinzhi. Road surface crack detection method based on deep learning[J]. Science Technology and Engineering, 2021, 21(15): 6380-6385.
[6]
Wang J, Gao X, Liu Z, et al. GSC-YOLOv5: an algorithm based on improved attention mechanism for road creak detection[C]// IEEE 12th Data Driven Control and Learning Systems Conference (DDCLS). Xiangtan: IEEE, 2023: 1664-1671.
[7]
王朕, 李豪, 严冬梅, 等. 基于改进YOLOv5的路面病害检测模型[J]. 计算机工程, 2023, 49(2): 15-23.
Wang Zhen, Li Hao, Yan Dongmei, et al. Pavement disease detection model based on improved YOLOv5[J]. Computer Engineering, 2023, 49(2): 15-23.
[8]
Wei C, Zhang Q, Zhang X, et al. YOLOv5s-BSS: a novel deep neural network for crack detection of road damage[C]// IEEE International Conference on Big Data (BigData). New York: IEEE Computer Society, 2023: 1550-1554.
[9]
罗晖, 余俊英, 涂所成. 基于深度学习的公路路面病害检测算法[J]. 科学技术与工程, 2022, 22(13): 5299-5305.
Luo Hui, Yu Junying, Tu Suocheng. Pavement defect detection algorithm based on deep learning[J]. Science Technology and Engineering, 2022, 22(13): 5299-5305.
[10]
倪昌双, 李林, 罗文婷, 等. 改进 YOLOv7 的沥青路面病害检测[J]. 计算机工程与应用 2023, 59(13): 305-316.
Ni Changshuang, Li Lin, Luo Wenting, et al. Disease detection of asphalt pavement based on improved YOLOv7[J]. Computer Engineering and Applications, 2023, 59(13): 305-316.
[11]
王保宪, 白少雄, 赵维刚. 基于特征增强学习的路面裂缝病害视觉检测方法[J]. 铁道科学与工程学报, 2022, 19(7): 1927-1935.
Wang Baoxian, Bai Shaoxiong, Zhao Weigang. Pavement crack damage visual detection method based on feature reinforcement learning[J]. Journal of Railway Science and Engineering, 2022, 19(7): 1927-1935.
[12]
Cao M T, Tran Q V, Nguyen N M, et al. Survey on performance of deep learning models for detecting road damages using multiple dashcam image resources[J]. Advanced Engineering Informatics, 2020, 46: 101182.
[13]
罗晖, 贾晨, 李健. 基于改进YOLOv4的公路路面病害检测算法[J]. 激光与光电子学进展, 2021, 58(14): 336-344.
Luo Hui, Jia Chen, Li Jian. Road surface disease detection algorithm based on improved YOLOv4[J]. Advances in Laser and Optoelectronics, 2021, 58(14): 336-344.
[14]
Iparraguirre O, Iturbe-Olleta N, Brazalez A, et al. Road marking damage detection based on deep learning for infrastructure evaluation in emerging autonomous driving[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23 (11): 22378-22385.
[15]
Arya D, Maeda H, Ghosh S K, et al. Deep learning-based road damage detection and classification for multiple countries[J]. Automation in Construction, 2021, 132: 103935.
2025年第25卷第2期
PDF下载
253
99
引用本文
BibTeX
文章信息
doi: 10.12404/j.issn.1671-1815.2401676
  • 接收时间:2024-03-11
  • 首发时间:2025-12-05
  • 出版时间:2025-01-18
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2024-03-11
  • 修回日期:2024-11-04
基金
广西信息材料重点实验室基金(221035-K)
中央引导地方科技发展资金(20231011)
广西高校中青年教师基础能力提升项目(2022KY0195)
作者信息
    1 桂林电子科技大学信息与通信学院, 桂林 541004
    2 广西无线宽带通信与信号处理重点实验室, 桂林 541004
    3 南宁桂电电子科技研究院有限公司, 南宁 530000

通讯作者:

* 陈紫强(1973—),男,汉族,湖南沅江人,博士,副教授。研究方向:信号处理。E-mail:
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/kxjsygc/CN/10.12404/j.issn.1671-1815.2401676
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数

Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)

Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏