Electricity Transmission Line Insulator Malfunction Detection Through Deep Learning

The transmission line is the most important part in the transmission of electric power.Insulators are used to connect wires and power transmission towers.Most of the insulators are made of ceramics,glass or composite materials.In order to enable the transmission line to operate normally,ensuring the integrity of the insulator has become an indispensable step for the maintenance of the transmission line.In recent years,with the continuous popularization of UAV technology,the use of UAV to patrol high-voltage transmission lines has become an important inspection method in line inspections.In the process of using UAV to photograph insulators,the background of insulators is complex,for example,they will be interfered by poles,towers,houses and other obstacles,which will bring great difficulties in the process of positioning and detection of insulators.At the same time,due to the enormous data taken by UAV,the staff will inevitably make great errors when judging with their naked eyes and experience,which will result in a great waste of manpower and material resources.Therefore,it is very important to perform target positioning and fault judgment on the obtained aerial image through computer vision assisting technology.This paper uses deep learning method to locate insulator and judge whether the insulator is broken,which has achieved good results.The article mainly carried out the following work:Firstly,this paper builds a commonly used convolutional neural network to realize the position detection of insulators.Based on GPU server,the algorithm of transmission line insulator detection is studied by using the method commonly used in deep learning.With Faster-R-CNN(Faster-Region--based Convolutional Neural Networks)algorithm for detection of insulator location,this method through the RPN(Region Proposal Network)regional candidate Network with Fast R-CNN(Fast Region--based Convolutional Network)to detect Network share whole image convolution feature,achieve end-to-end testing;Although the robustness and generalization ability of the algorithm are improved compared with the traditional algorithm,the accuracy rate fails to meet the ideal requirements.Therefore,a series of experiments were conducted to find the optimal number of network layers and feature map,and feature fusion of multiple convolutional layers was conducted to improve the accuracy of convolutional neural network in insulator positioning.Secondly,on the basis of the insulator detection,through fine tune of the pre-trained convolution network,the judgement of whether the insulator has fallen off is completed.The computer aided the judgment of the fault,which solved the problem that the workload of the manual search for the insulator was too large.During the training of convolutional network,the fault judgment ability of the network model for insulators finally reached about 98.6%by adjusting the super-parameters such as the learning rate and dropout of neural network.Most of the previous insulator data sets were not made public.The author marked the position of 500 insulators by himself,manually classified 13000 insulator images,and expanded the data.In the process of debugging the network,step-by-step debugging,doing batch identification and discrimination of Insulator Integrity were carried out.