改进YOLOv5的输电线路无人机图像缺陷检测研究  

Research on Improving YOLOv5 for Detecting Defects in Unmanned Aerial Vehicle Images of Transmission Lines

作　　者：周霞[1] 揭振华 朱兆优[3] 揭博程 ZHOU Xia;JIE Zhenhua;ZHU Zhaoyou;JIE Bocheng(Jiangxi Construction Vocational and Technical College,Jiangxi Nanchang 330200,China;Jiangxi Quench Information Technology Co.,Ltd.,Jiangxi Nanchang 330038,China;East China University of Technology,Jiangxi Nanchang 330200,China)

机构地区：[1]江西建设职业技术学院,江西南昌330200 [2]江西淬火信息科技有限公司,江西南昌330038 [3]东华理工大学,江西南昌330200

出　　处：《机械设计与制造》2025年第3期262-266,273,共6页Machinery Design & Manufacture

基　　金：江西省教育厅科学技术研究项目(GJJ205304)。

摘　　要：针对现有无人机航拍输电线路图像缺陷检测方法存在的检测精度差和效率低等问题,提出一种用于输电线路图像缺陷检测的改进YOLOv5模型。通过四个方面的优化(Kmeans+算法、SENet注意力模块、DIoU损失函数、优化空间金字塔化结构)提高YOLOv5模型的检测精度和效率。通过实验验证了所提模型在绝缘子缺陷和杆塔鸟巢缺陷检测中的优越性。结果表明,与常规输电线路缺陷检测方法相比,所提方法具有较高的缺陷检测准确率和较快的检测速度,检测准确率大于95.00%,检测速度高于40FPS。该研究可为电力巡检提供一定的帮助。An improved YOLOv5 model for defect detection of transmission lines is proposed to address the issues of poor detection accuracy and low efficiency in existing unmanned aerial vehicle aerial image defect detection methods.The detection accuracy and efficiency of YOOv5 model are improved through four aspects of optimization(Kmeans+algorithm,SENet attention module,DIoU Loss function,optimization of spatial pyramid structure).The superiority of the proposed model in detecting insulator defects and tower bird's nest defects was verified through experiments.The results show that compared with conventional transmission line defect detection methods,the proposed method has higher defect detection accuracy and faster detection speed,with detection accuracy greater than 95.00%and detection speed higher than 40FPS.This study can provide certain assistance for power inspection.

关 键 词：无人机 输电电路 图像缺陷 YOLOv5模型 SENet注意力模块 

分 类 号：TH16[机械工程—机械制造及自动化]