Transmission Line Insulator Identification And Defect Location Based On YOLOv4 Algorithm

With the development of the economy and the rapid increase of electricity load,the range of transmission lines is increasing and the task of line detection is becoming more and more difficult.The traditional manual inspection method is time-consuming and inefficient,and the fault detection rate is low.The maturity of artificial intelligence technology,the use of unmanned aircraft carrying high-definition equipment for inspection is becoming more and more widespread.Insulators in transmission lines are the most numerous and easily damaged components,and their main role is to insulate and support,so regular inspection of insulators on transmission lines is necessary.Therefore,this paper integrates power inspection with vision processing technology to identify and locate faults in insulators on transmission lines.To achieve insulator identification using UAVs with movable devices,this paper adopts an improved target detection network(K-means-Mobile Netv2-YOLOv4)for insulator identification under deep learning.To improve the detection speed and reduce the model size,this paper introduces a lightweight deep separable network(Mobile Net)to replace the backbone extraction network of YOLOv4.Three versions of Mobile Netv1,Moblie Netv2,and Moblie Netv3 are selected as the backbone extraction network of YOLOv4 for insulator detection experiments,compared with the other The two versions Mobile Netv2-YOLOV4 identify insulators with the highest accuracy,but the identification accuracy is not as good as the original YOLOv4 algorithm model.In this paper,K-means clustering reset frame is introduced to solve the situation of low insulator recognition accuracy in this paper.The algorithm is improved by conducting experimental comparison of recognizing insulators under different targets in various environments,and then comparing the performance with various mainstream algorithm models.Secondly,the improved target detection network is used to locate defects in insulator self-explosion,and the disadvantages of the traditional way of detecting insulator defects are used as a direction to lead to the improved algorithm model for identifying insulator defects in this paper,which first expands the insulator defect data set,trains the defect images using Mosaic data to achieve the best detection effect,and then introduces the SE attention mechanism to further improve the detection defect The accuracy of the detection is further improved by introducing the SE attention mechanism.The experimental results show that the K-means-Mobile Netv2-YOLOv4 algorithm model identifies insulators with 91.81% accuracy,52 MB model size,and 46 FPS detection speed,which has a loss in accuracy compared with the YOLOv4 algorithm model,but has an improvement in model size and detection speed.Compared with Faster R-CNN,SSD,and YOLOv3 algorithm models,Map is improved by 3.69%,5.02%,and 2.43%,respectively;in insulator defect detection,the detection effect is better compared with traditional methods more,and the accuracy of insulator defects reaches 90.78% after the introduction of SE attention mechanism,and the improved insulator defect localization network detection model can guarantee the The improved insulator defect localization network detection model can realize the detection of transmission line insulators with the accuracy rate.