| The Catenary inspection is related to the safety of railway operation.In order to ensure the stable operation of the railway,the railway corporation has developed the railway power supply system safety monitoring platform(6C system),the catenary suspension monitoring system(4C system)detects each part of the catenary by the method of manual detection and traditional template matching.However,the traditional template matching and manual detection methods have some problems such as poor generalization ability and need to rely on a lot of manual work.In recent years,the computer vision technology based on deep learning has developed rapidly,so this paper combines the new algorithm of deep learning object detection to study the theory and experiment of insulator detection technology.The insulator dataset is constructed by collecting and sifting a large number of on-site images of overhead contact lines.The insulator images of different states are selected according to the need,and then the insufficient number of damaged samples are expanded by data enhancement.All insulator images are manually labeled using LABELIMG software,and the dataset is labeled and stored as a VOC dataset for training and testing the network model.In order to solve the problems of low speed and low real-time performance of the fault detection method for railway insulators,an improved method based on Yolov4-tiny is proposed.At first,three different convolution attention modules are introduced respectively,and the experiments are compared and analyzed.Then,the convolution attention module which has the best experimental effect is added to improve the target detection network,finally,the experiment is compared with other classical target detection algorithms.The experimental results show that this method can make the network adaptively extract the effective features of the target object,improve the detection speed and be suitable for real-time detection.In order to solve the problem that the precision of fault detection method of railway insulators is not ideal,an improved method based on efficient det is proposed.Firstly,the Kmeans algorithm is used to recluster the samples in the insulator data set to generate a more targeted anchor,and then,in the contact network insulator breakage detection model,the feature fusion module is redesigned,and the accuracy of fault classification in insulator data set is verified by the comparison experiment.The experimental results show that the problem of detecting small targets can be effectively reduced by reclustering anchor frames,and the efficiency of feature fusion can be effectively improved by feature fusion optimization,the precision of insulator detection is further improved.The two detection algorithms designed in this paper can locate the target insulators accurately and classify the insulators correctly.The improved YOLOv4-tiny detection algorithm outperforms in the detection speed,and the improved efficientdet detection algorithm outperforms in the detection precision. |
