基于YOLOv5的无人机桥面病害检测算法研究

打开文本图片集

DOI：10.16652/j.issn.1004-373x.2025.16.002

中图分类号：TN911.23-34；TP391.4 文献标识码：A 文章编号：1004-373X（2025）16-0007-06

ResearchonUAVbridgedeck disease detectionmethod based on YOLOv5

DAI Pengfeil，²， ZOU Jingshan³4，YANG Liu3，4 ，LIU Heng3，4 ， YIN Huiying4， 5

（1.NanjingchUivesityg16a;2.inlagedugod.g; 3.School of Information Science and Technology，Southwest JiaoTong University，Chengdu 611756，China; 4li 5.Tangshan Institute，Southwest JiaotongUniversity，Tangshan O63ooo，China）

Abstract：Bridge diseases，suchasconcretepeeling，bridge cracks，rivetcorosion，etc.，mostlyoccur inlocalareas，but mostof thebridgediseasesarenotlocatedinthewhole bridgeatpresent.Theful-fieldrapidlocationanddetectionof bridge deckdiseasescanberealizedbycombining theigh-definitioncamerafunctionofunmannedaerialvehicle（UAV）withtherealtimetargetdetectionabilityof YOLOv5.Therefore，an UAVbridgedeck disease detectionalgorithm basedonYOLOv5is proposed.The UAVisused tocolectdataon thebridge pavement，andthelightweight modelYOLOv5sisusedasthebasic detectionmodel.TheYOLOv5smodelisimprovedasfolows：twoscalesareaddedonthebasisoftheexisting three characteristicmapsdetectionwithdiffrentscalestoimprovethedetectionaccuracyoflargertargetsandsmaertargets;SoftNMSalgorithmisusedtoinsteadofNMSalgoritm.Inordertoensuretheful-fieldrapidpositioninganddetectionaccuracyof densediseases，thecollctedbridgepavementdataisinputintotheimprovedYOLOv5smodel，andtheoutputofthemodelis the detection result of bridge deck diseases.The experimental results show that the value of mAP of the optimized YOLOv5s model can reach 92.0% ，and the value of mAP@0.5：0.95 also can reach 73.2% . The processing speed of the model can reach 134f/s，whicheffectivelyandacuratelyidentifiesbridgepavementdiseases，andsignificantlyimprovestheaccuracyand efficiency of detection.

Keywords：bridge deck disease detection；YOLOv5;unmannedaerial vehicle;image colection;multi-scale detection; feature fusion

0 引言

桥梁作为轨道交通的重要组成部分，其建设在轨道交通修筑中占有重要地位。（剩余6414字）