Research On The Method Of Recognizing Typical Defects In GIS Based On Image Features

Gas-insulated switchgear(GIS)is an important high voltage electrical device in UHV power grid,and it is the main switch equipment in the power grid.GIS equipment is flexible,safe,easy to maintain and long maintenance period,has been widely used in the power system However,with the large-scale operation of GIS equipment,some defects are gradually exposed,which seriously affects the normal operation of the power grid.So how to accurately detect GIS devices has become the most important issue now.First,through the case analysis of typical defects of GIS equipment,the most frequent and most common insulation faults are divided into 3 types: protrusion defect,free metal particle defect and surface crack.By studying the characteristics of three kinds of fault images and normal state images,we analyze and calculate 5 texture features and 6 shape features of the images,and select a deficit moment and contrast,2 texture features and 6 image invariant moments to identify the feature vectors of the image.On this basis,image classification method based on BP neural network and image classification method based on convolution neural network are given respectively.The test results verify the effectiveness of the proposed method.Then,the GIS typical defect image recognition system is developed,based on Matlab,which includes three modules: defect image recognition,defect image management,and update neural network.The main function of the defect image recognition module is added and upload the defect image,showing the recognition results in the interface;the main function of defect image management module is to add,delete,defect image defect images add and delete the defect type defect type;main function update neural network module is updated for neural network image or defect the type of defects.