电瓷避雷器Issue(6)：145-152,8.DOI:10.16188/j.isa.1003-8337.2024.06.017
融合注意力与轻量化的多尺度电缆绝缘子缺陷检测
Multi-Scale Defect Detection of Cable Insulators Based on Attention and Lightweight
摘要
Abstract
Aiming at the problems of low detection accuracy and complex network of existing insulator de-fect detection methods,a multi-scale cable insulator defect detection method that integrates attention and lightweight is proposed.Firstly,channel attention is introduced into the YOLOv5 backbone network,suppressing complex background texture and emphasizing defect feature information.Secondly,the Ghost-Bottleneck module is designed based on the lightweight characteristics of the Ghost module,and the Bottleneck module in YOLOv5 is replaced,which reduces the complexity of the network.Then,the K-means++algorithm is used to cluster the anchor box coordinates of the prediction box,and the multi-scale coupling head is added and aggregated to improve the detection accuracy of small defect targets.Fi-nally,the DIOU Loss positioning loss function is modified to the CIOU Loss positioning loss function to optimize the network parameters.Test results show that the mAP of insulator defect detection on the CPLID dataset reaches 99.5％,and the number of parameters is reduced by 33.5％compared with the original YOLOv5 network.The algorithm realizes the detection of insulator defects and is easier to trans-plant to the inspection terminal,which provides a new and more accurate method for intelligent inspection of insulator defects.关键词
目标检测/绝缘子缺陷/YOLOv5/多尺度/Ghost模块/注意力机制Key words
object detection/insulator defects/YOLOv5/multi-scale/Ghost module/attention mecha-nisms引用本文复制引用
赵洋,刘青,李宁,尚英强,马宪伟..融合注意力与轻量化的多尺度电缆绝缘子缺陷检测[J].电瓷避雷器,2024,(6):145-152,8.基金项目
国家电网有限公司科技项目(编号:5500-202111118A-0-0-00).Project supported by Science and Technology Program of State Grid(No.5500-202111118A-0-0-00). （编号:5500-202111118A-0-0-00）
