Research On Power Transmission Equipment Detection Technology Based On Image

In the past ten years,with the vigorous development of the power industry in China,our country has created many world firsts in the field of transmission lines and exported transmission technology and equipment abroad,which achieving "Made in China" and making transmission lines became one of China’s new business cards and led the world on behalf of China.Transmission lines have a very important position in all areas of society,and the operating status of many transmission equipment in the transmission line directly affects the stability and safety of the power system.Therefore,it is necessary to conduct regular inspections on transmission lines,especially the wires and tower equipment in the transmission lines.At present,the main methods of transmission line inspections are manual inspections,drone inspections,etc.Manual inspections require inspectors to climb to the transmission line and rely on manual naked eye observation,which is time-consuming,laborious and the inspection results are greatly affected by subjective factors.And the location of the overhead transmission lines is special,which passes through farmland,mountains,rivers,etc.,bringing a certain degree of danger to the personal safety of inspectors.At present,the current drone inspections mostly require manual assistance.The videos and images captured by the pan/tilt camera are stored in the SD card and transmitted back to the ground in real time for the inspectors to observe the operating status of each transmission equipment.The degree of intelligence is low.Aiming at the above problems,this paper studies the image detection methods of wires and towers.The convolutional neural network trains and learns the input training set,continuously optimizes and adjusts the weight parameters,and automatically extracts target features.It overcomes the traditional digital image processing technology that only uses the target texture,line segment,and gray value.Such shallow features result in low detection accuracy,and require artificial design of target features,and the calculation process is cumbersome.Therefore,this paper is based on the convolutional neural network to detect the wires and towers in transmission line image.According to the respective characteristics of the wires and the towers in the image,the wire detection based on the convolutional neural network is equal to the task of semantic segmentation of the wires.The pole-tower detection based on convolutional neural network boils down to the target detection of the tower.The main research work and research results are as follows:First,this paper studies the transmission line detection method based on semantic segmentation network.By analyzing the characteristics of the wires in the transmission line image,the DeepLab v3+network is used to realize the segmentation of the transmission wires in the image.In view of the pseudo wires and the discontinuity of single wire in the wire segmentation results,a refined processing scheme is designed to output the complete wire detection results.In addition,on the basis of the DeepLab v3+network,the paper improves DeepLab v3+ network from the feature extraction network and the ASPP structure and proposes MDeepLab v3+network,which can effectively increase the speed of wire segmentation while ensuring the accuracy of wire segmentation and solve the problem of the slow detection speed of DeepLab v3+network in actual application.Second,this paper studies the detection method of power transmission towers based on the target detection network.First,the one-stage target detection network YOLOv3 is used to identify and locate the position of the tower in the image;and under the premise of ensuring the detection accuracy,by studying the structure and parameter settings of the YOLOv3 network,in order to improve the detection speed of the algorithm,the network has been lightened and improved from the two parts of the feature extraction network and the multi-scale detection network Based on the large and small versions of the MobileNetV3 network,the paper studies four light-weighted improvement methods of YOLOv3,and proves through experiments that these four light-weighted improvement methods all can help improve the detection speed of the network.One of the light-weighted improvements uses the large version of MobileNetV3 as the feature extraction network of YOLOv3,while modifying the number of channels of each convolutional layer in the YOLOv3 multi-scale detection network.This improvement method can detect video in the NVIDIA Jetson Xavier NX device at the speed of 22 frames/s,which is more than 3 times higher than that of YOLOv3 network,and the detection accuracy is only reduced by 1.1%.