Insulator Recognition Based On Improved Convolution Neural Network

Insulator is an important equipment in transmission line,its operation status directly affects the stability and normal operation of power system.Therefore,it is necessary to use intelligent methods to process and analyze a large number of insulator image data acquired by UAV,so as to provide reliable reference information for power inspection.Aiming at the task of insulator fault classification,this paper studies from the following two aspects:(1)The premise of insulator fault classification is to accurately identify and locate the insulator on the insulator image.Therefore,in this paper,firstly,the insulator recognition algorithm based on convolution neural network is studied.Aiming at the problems of inaccurate insulator target location and easy to miss insulator detection with smaller size using Faster RCNN target detection model,an insulator recognition method combining attention mechanism and Faster RCNN is proposed.First,the Squeeze-and-Excitation Networks structure is introduced into the VGG16 network to filter and enhance the characteristic channel related to insulator target and weaken other target independent channels.Secondly,adjust the generation mechanism of anchor points(anchor)in the region proposal network(RPN),accelerate the convergence speed of the model,and improve the recognition accuracy.Finally,the attention mechanism is used to fuse and update the feature vectors of the suggestion frame on the full connection layer.The experimental results show that compared with the traditional Faster RCNN target detection algorithm,the improved algorithm can identify and locate the insulators of different scales better,with higher recognition accuracy and lower missing detection rate.(2)In order to further realize the task of insulator fault classification,this paper proposes an insulator fault classification model based on the improved convolution neural network insulator recognition algorithm and image processing algorithm.Firstly,the identified insulator target area is clipped to remove a lot of background noise;secondly,the HSV color space adaptive threshold algorithm and image horizontal correction are used to deal with the influence of light,background noise and shooting angle in the clipped image;finally,the adaptive threshold segmentation and vertical projection are used to classify the fault of the segmented insulator.The experimental results show that the insulator fault classification model proposed in this paper can effectively identify the insulator and classify the insulator fault types,which can meet the requirements of intelligent inspection of transmission lines,and has high engineering application value.