首页|基于改进YOLOv5s的输电线路螺栓缺销检测方法

基于改进YOLOv5s的输电线路螺栓缺销检测方法

扫码查看
原文链接
  • 万方数据
  • 维普
针对无人机输电线路巡检图像中螺栓缺销检测精度较低、漏检较多的问题,提出了一种基于改进YOLOv5s的输电线路螺栓缺销检测方法.在Backbone部分嵌入Coordinate Attention注意力模块;在Neck部分原有的"FPN+PAN"结构的基础上,新增一条"自顶向下"的特征信息传递路径,跨越临近的同尺度特征层,与较浅层网络以加权融合的方式进行特征融合;将Head部分设置为解耦检测头,将对螺栓检测的分类任务与定位任务分开进行.改进后的YOLOv5s算法增强了对螺栓特征信息的学习能力.使用本方法在螺栓缺销数据集上实验,精确率提升了2.3％,召回率提升了3.4％,平均精度提升了3.1％,检测速度达到了41.1帧/秒,表明改进后的方法能提升输电线路螺栓缺销的检测能力,在智能巡检中具有一定的应用价值.
Detection Method for Pin-losing-bolts in Transmission Lines Based on Improved YOLOv5s
Aiming at the problems of low detection accuracy and many missed inspections of bolts in the inspection ima-ges of drone transmission lines,we proposed the detection method for pin-losing-bolts in transmission lines based on improved YOLOv5s.In the Backbone part,the Coordinate Attention module is embedded.Based on the original"FPN+PAN"structure of the Neck part,a"top-down"information transmission path is added,which spans the adja-cent feature layer of the same scale,and features are fused with the shallower network in the way of weighted fusion.The Head part is improved into a decoupled head,which separates the classification task of bolt detection from the lo-calization task.The improved YOLOv5s algorithm enhances the learning ability of bolt feature information.Using this method to experiment on the pin-losing-bolts dataset,the accuracy rate was increased by 2.3％,the recall rate was in-creased by 3.4％,the average accuracy was increased by 3.1％,and the detection speed reached 41.1 frames/sec-ond.It shows that the improved method can improve the detection ability of pin-losing-bolts in transmission line,and has certain application value in intelligent inspection.

patrol imagefault detectionpin-losing-boltsYOLOv5sCoordinate Attentionfeature fusiondecou-pled head

赵文清、贾梦颖、翟永杰、赵振兵

展开 >

华北电力大学控制与计算机工程学院,河北保定 071003

复杂能源系统智能计算教育部工程研究中心,河北保定 071003

华北电力大学电气与电子工程学院,河北保定 071003

巡检图像 故障检测 螺栓缺销 YOLOv5s Coordinate Attention 特征融合 解耦检测头

国家自然科学基金河北省自然科学基金中央高校基本科研业务费专项面上项目中央高校基本科研业务费专项面上项目

61871182F20215020132020MS1532021PT018

2024

10.3969/j.ISSN.1007-2691.2024.03.11

华北电力大学学报(自然科学版)
华北电力大学

华北电力大学学报(自然科学版)

CSTPCD北大核心
影响因子：0.868
ISSN：1007-2691
年,卷(期)：2024.51(3)
  • 12