Insulator Visual Recognition Technology Based On Multi Kernel Function SVM

It is of vital importance for eliminating hidden danger of pollution flashover and ensuring the stable operation of the railway transportation system to study visual insulator identification system and realize intelligent cleaning of insulators of the railway contact network.In this thesis,visual insulator identification models were established with kernel function based support vector machine algorithm(SVM),and the identification and detection of insulator image samples was realized.The main research contents are as follows:(1)Insulator image samples were collected for binaryzation and ROI area extraction,and insulator feature samples were established through texture feature describing operators like Hu,LBP,HOG,Haar,etc..Training and test sample sets were set up with normalization and principal component analysis method.(2)Linear kernel,radical basis kernel and polynomial kernel based SVM training models and single texture type training model of linear kernel-based SVM method were built,and through a comparative analysis on the time spent and the identification performance of each model,the effectiveness of kernel function based SVM method on insulator identification was verified and the multi-core form of combination of visual insulator identification models was determined.(3)The influence of parameters on the performance of the final identification model in the kernel function-based SVM method was analyzed,the model identification accuracy rate and the number of support vectors were taken as the comprehensive indexes for evaluation of model performance,and the corresponding evaluation function was brought up.On the basis of the analysis on a variety of swarm intelligent optimization algorithms,parameters of kernel function based SVM method were optimized with differential evolution algorithm and the performance of the insulator identification model was further improved.(4)Hexagonal lattice calibration plate and SVM prediction algorithm were adopted to build an insulator space positioning system with whole horizon sampling and binocular vision,and the effectiveness of the model was verified through an analysis on the space positioning accuracy of the mapping model by experiment.