基于改进YOLOv5的输电线路缺陷检测方法

Defect detection method of transmission line based on improved YOLOv5

目前,电网公司逐渐采用无人机对电气设备进行智能巡检,其中绝缘子缺陷、R销缺失和鸟巢检测是重要的巡检环节。提出了一种改进YOLOv5算法,以实现输电线路缺陷检测。首先,对无人机采集的超高分辨率图片进行切图处理,再利用K-means算法对数据集进行分析以获得最佳尺寸的锚框;其次,采用双向特征金字塔网络(bidirectional feature pyramid network,BiFPN)替换原先YOLOv5的特征金字塔网络,实现更加自由的信息交换和特征融合;最后,添加SimAM注意力模块来解决复杂背景对缺陷识别精度的影响。试验结果表明:改进后的YOLOv5算法检测效果显著提升,具有较高的应用价值。

Currently,power grid companies are gradually adopting unmanned aerial vehicles(UAVs)for intelligent inspections of electrical equipment,with a focus on detecting insulator defects,missing R cotter,and nesting.This paper proposes an improved YOLOv5 algorithm to enable defect detection in transmission lines.Firstly,the ultra-high-resolution images captured by the UAV are processed into smaller patches,and the dataset is analyzed using the K-means algorithm to obtain the optimal anchor box size.Secondly,BiFPN(bidirectional feature pyramid network)is adopted to replace the original feature pyramid network of YOLOv5 to realize more free information exchange and feature fusion.Finally,the SimAM attention module is added to solve the influence of complex backgrounds on defect identification accuracy.The experimental results show that the detection effect of the improved YOLOv5 algorithm is significantly improved,so it has high application value.