改进YOLOv5s算法的小目标输电导线缺陷检测方法

打开文本图片集

关键词：无人机巡检；输电导线；缺陷检测；卷积块注意力模块YOLOv5；交并比损失DOI：10.15938/j. jhust.2025.05.004中图分类号：TP751.1 文献标志码：A 文章编号：1007-2683（2025）05-0041-09

Abstract：Asthecarrierof transmision lines，power transmissionconductorsaresusceptible tothe influenceof theexternal environmentandtheoccurenceoflinedefectssuchasbrokenconductorstrandsorloseconductorstrands.Aimingattheissuesof poordetectionperformancecausedbytheinfluenceofcomplexbackgroundsduringunmanedaerialvehicle（UAV）cruising inspectionsofsisilies，swellassdandfleectiosuetcessielysallfoductordectets aerial images，a detection method named CBAM YOLOv5 based on the atention mechanism is proposed. Firstly，the original attntion mechanismCBAMfeatureextractionisoptiizedtoincreasetheextractionefectof transmissonconductorfeatureinformationAthe samtime，themulti-scaletrainingstrategyisusedtoimprovethemodel'sabilitytoperceivethefeatureinformation.Finally，ore superiorlossfunction WIULossisproposedtocopewiththeproblemofunbalanced positiveandnegativesamplesinthesmaltarget. Theexperimentalsimulationshowsthattheimprovedalgorithmappliedonthewiredefectdatasetimprovesthecomprehensive performance by O. O67 compared with YOLOv5s，and the accuracy reaches 87.9% ，which is 2% higher than YOLOv5s，verifying the effectiveness of the improved algorithm.

Keywords：unmannedaerialvehicle inspection；transmisionconductor；defectdetection；convolutional block atentionmodule YOLOv5 ；Wise-IoU Loss

0引言

截至2015年底，中国220千伏及以上输电线路总长60.9万公里。（剩余11735字）