基于混合注意力机制的YOLOv5s防振锤检测方法  

Vibration damper detection method based on mixed attention mechanism YOLOv5s

作　　者：唐锐 陈剑波 姚平 张赛飞 廖林 杨春萍[2] TANG Rui;CHEN Jian-bo;YAO Ping;ZHANG Sai-fei;LIAO Lin;YANG Chun-ping(State Grid Xinjiang Electric Power Co.,Ltd.Bazhou Power Supply Company,Korla 841000,Xinjiang Uygur Autonomous Region,China;School of Electrical and Electronic Engineering,North China Electric Power University,Beijing 102206,China)

机构地区：[1]国网新疆电力有限公司巴州供电公司,新疆库尔勒841000 [2]华北电力大学电气与电子工程学院,北京102206

出　　处：《信息技术》2025年第4期8-15,共8页Information Technology

基　　金：国网新疆电力有限公司科技项目(5230BD220001)。

摘　　要：针对输电线路无人机巡检中防振锤尺度差异大、背景复杂和防振锤分布集中导致检测精度不佳的问题,提出一种基于混合注意力机制的YOLOv5s防振锤检测方法。首先通过学习不同通道重要程度来提升二阶通道注意力机制对目标的关注能力,增强模型对防振锤的检测效果,同时基于Swin Transformer自注意力机制优化YOLOv5s的检测输出,提高对密集目标的关注能力。为使预测框更加准确逼近真实框,加强对目标的定位能力,引入SIoU函数提升边框回归的精度。实验结果表明,所提方法检测精度较YOLOv5s模型提高了3.6%,检测速度达到41帧/秒,证明该方法具有较好的实时检测能力。Based on the problem of poor detection accuracy caused by large differences in vibration damper scale,complex background and concentra ted vibration damper cloth in UAV inspection of transmission lines,a vibration damper detection method based on mixed attention mechanism YOLOv5s is proposed.Firstly,by learning the importance of different channels,the attention mechanism of second-order channels can be improved to enhance the target attention ability of the model to enhance the detection effect of the vibration damper.Meanwhile,based on Swin Transformer self-attention mechanism,the detection output of YOLOv5s can be optimized to improve the focus ability of dense targets.I n order to make the prediction frame more accurately approximate the real frame and strengthen the localization ability of the target,SIoU function is introduced to improve the accuracy of the frame regression.Experiment results show that the detection accuracy of the proposed method has gain 3.6%improvement than YOLOv5s model,and the detection speed is up to 41 frames per second,which proves that the proposed method has good real-time detection ability.

关 键 词：输电线巡检 YOLOv5s 注意力机制 SIoU 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]