Insulator Detection And Defect Recognition In Transmission Line Patrol Image Based On Deep Learning

With the continuous expansion of the EHV transregional large capacity transmission lines in China,how to ensure the safe operation of transmission lines and timely eliminate safety risks has become one of the main problems faced by inspection personnel.At present,the UAV inspection of transmission lines is gradually replacing the traditional manual inspection due to its advantages of safety,reliability and low cost.With the large-scale application of UAV inspection,how to conduct intelligent,efficient and accurate detection and analysis of massive inspection images through computer vision technology is one of the research hotspots of realizing intelligent electric power inspection system.In this paper,the insulator,the main power device of transmission line,is taken as the detection object to construct the insulator data set based on PASCAL VOC2007 format and COCO format.According to the inspection task’s requirements for detection accuracy and speed,the insulator detection algorithm based on YOLOV3 and CenterNet were put forward respectively to detect the insulators and the self-breaking defects in the inspection images of transmission lines.The main work is as follows:To solve the problem of insufficient insulator image data for research,an expansion method of insulator image data based on the Albumentation image enhancement framework is proposed,which enriches the insulator data set and improves the problem of low network detection accuracy caused by insufficient data.In order to meet the requirements of high precision and high speed of power inspection,a method of insulator detection and defect identification based on anchor frame network YOLOV3 is proposed.This detection method combined with spatial pyramid pooling structure can enrich the expression ability of feature map.K-means++clustering algorithm is used to redesign the size of anchor point frame,which makes the anchor point frame match the actual size of insulator,and improves the accuracy of positioning algorithm.Experimental results show that the improved algorithm improves the detection accuracy while maintaining a high detection speed,and can effectively identify insulators and insulator self-rupture defects.Considering the problems that anchor box detection network relies too much on manual design of anchor box size and the imbalance of positive and negative samples,a new insulator detection method based on CenterNet of anchor box detection network is proposed.Resnet34 is used as the feature extraction network to identify the insulators by detecting the center point and boundary information of the insulators in the inspection image.Under the premise of ensuring the detection accuracy,the detection method has simple network structure and less memory requirement for network operation,and is easy to be deployed in the field.Therefore,it can be applied to online inspection and has a certain engineering application prospect.