Research On Insulator And Defect Detection Algorithm Based On Deep Learning

The transmission line is an important carrier for transmitting electricity,and its safe and stable operation cannot be separated from the coordination of each component.Insulators are important components for electrical insulation and wire fixation on transmission line towers,and long-term exposure to natural environments leads to frequent faults.Regular maintenance of insulators is of great practical significance to ensure normal transmission,avoid risks and losses.This article focuses on the problems of low number of insulators,diverse sizes,and small burst areas in aerial insulator images.Based on the YOLOv5 s and YOLOX-S object detection algorithms,research is conducted on the detection of normal and self-detonation insulators.The details are as follows:(1)A standard insulator dataset PLI-Insulator has been established.This article screened and classified the transmission line dataset provided by Guangxi Jinghang Unmanned Aerial Vehicle Co.,Ltd.to obtain a small number of insulator images containing self-detonation defects.After ordinary data augmentation and manual annotation,the insulator dataset PLI-Insulator was created according to the PASCAL VOC2007 dataset format.(2)Designed an insulator and its defect detection algorithm based on YOLOv5 s.Firstly,after the first convolutional layer of the YOLOv5 s backbone,a convolutional block attention module(CBAM)is introduced,so that the model focuses on the key information of insulators in the initial feature extraction stage.In addition,in order to make the improved convolution layer have more abundant performance capability,the original activation function of the convolution layer is replaced by Hard Swish to ensure that the convolution attention module has a stable input.Finally,replace the activation function of other convolution layers in the network hidden layer with a smoother Mish,so that the input information can flow into the deep layer of the network better.The experimental results show that compared with the original YOLOv5 s algorithm,the improved algorithm has an average detection accuracy improvement of 7.8% on the PLI-Insulator dataset.(3)Designed an insulator and its defect detection algorithm based on YOLOX-S.Using YOLOX-S with an anchor free frame strategy as the basic algorithm,in order to improve the channel weight of insulator features during feature extraction,an efficient channel attention mechanism(ECA)is introduced after each Bottleneck CSP structure in the backbone.At the same time,in order to integrate more insulator features during feature stacking,the ECA mechanism was introduced into each upsampling layer and convolutional downsampling layer of the Neck section to enhance the insulator features in the sampling results.In addition,the positioning loss function of the Prediction part is replaced by the CIo U Loss function with more accurate positioning and faster convergence speed,so as to optimize the network training results.The average detection accuracy of the designed algorithm on the PLI-Insulator dataset has been improved by 5.2%.