Research On Infrared Image Faults Recognition For Electrical Equipments Based On Dual Supervision Signals Convolution Neural Network

With the development of infrared imaging technology,infrared detection technology,using in thermal fault diagnosis of electrical equipments,has become one of the mainstream method in electric systems inspection.In order to improve the intelligentization of power system,and solve precise thermal faults detection problem in asubstation electrical equipments.This thesis puts forward a method of infrared image faults recognition for electrical equipments based on dual supervision signals convolutional neural network(CNN),which is one of known deep learning technology.In preprocessing stage of fault equipment image,according to the characteristics of infrared image imaging,a HSV space transform algorithm based on Slic super-pixel segmentation is proposed to locate the fault areas.The slic super pixel segmentation algorithm is adopted to merge the similar pixel regions into blocks.According to the luminance information provided by the improved HSV space transformation,the temperature abnormal regions are determined,then the connected area and the corresponding device of this region are separated.In identification stage,an algorithm that based on the GoogLeNet convolution neural network model is proposed.Fault features of infrared images for electrical equipments are extracted.A novel training and supervised method of extracted features is proposed,which is weighted mixed two kinds of signals,i.e.,the softmax loss and the center loss signal.In order to evaluate the performance of the method,we established an 700 images dateset of infrared fault of electrical equipments,500 of which are for network training,and 200 for experiment testing.Experiments results show that the test accuracy rate can reach to 98.6% which enhanced1% compared with the classic method that simply using the single softmax loss.The algorithm can accurately locating and recongnition ten kinds of common electrical equipment which include the transformer bushing,current transformer,insulator,lightning arrester,isolating switch,transformer,circuit breaker,cable,overhead line conductor,transformer cooler,as well as identify the corresponding faults.