Research On Detection And Recognition Method Of Pantograph Arc Based On Machine Learning

The pantograph is an important equipment for trains to obtain and transmit electric energy from the catenary.When the pantograph is in contact with the catenary,it is easy to cause arcing.In severe cases,it will cause harm to train equipment and catenary equipment,threatening the train.Safe operation.In view of the low detection efficiency of the existing arc detection methods that mostly use camera and manual analysis,by studying a large number of domestic and foreign arc detection documents,this thesis is based on traditional and deep learning target detection and recognition algorithms to draw the pantograph.Conduct testing research.The specific work is as follows:(2)Aiming at the problem of low detection rate of single feature extraction for arc recognition,a pantograph arc recognition model based on improved feature fusion SVM is designed.The color moment,GLCM,and HOG feature extraction algorithms are used to extract the arc feature information,and the extracted features are fused with a weighted fusion algorithm based on the accuracy of single feature recognition,and then input to the classifier SVM training,and the pantograph is realized through the classifier Recognition of arced images.Experiments show that the feature fusion algorithm improves the accuracy of pantograph arc recognition compared with single feature and direct serial fusion,and the accuracy rate can reach 89.4%.(2)Aiming at the slow speed of pantograph arc detection,a lightweight MobileNet-SSD pantograph arc detection model is designed,and the loss function and activation function of the model are improved.The SSD arc detection with good detection accuracy and speed is selected.In order to make the model small and real-time and easy to apply in the train,the basic network of the SSD model is replaced with a lightweight MobileNet network,and the MobileNet-SSD arc detection model is designed In order to improve the detection accuracy and reduce the imbalance of positive and negative samples,the activation function and loss function are improved.The average detection accuracy of the input image size of the arc is81.29%,and the FPS is 45 frames/s,which is 13 frames/s higher than the detection speed of the SSD algorithm.(3)Aiming at the problem that the MobileNet-SSD algorithm has low detection accuracy for small arcs,an improved algorithm based on MobileNet-SSD is proposed.The shallow Conv11 and Conv13 are introduced into the feature fusion enhancement module to enhance the shallow feature information and improve The ability to detect smaller targets.Experiments show that the average detection accuracy of the improved algorithm is 87.99%,and the detection speed FPS is 39 frames per second.The detection accuracy is greatly improved when the detection speed is limited.(4)Comparative analysis of the performance of the arc detection methods.The proposed arcing detection algorithm is used to detect the arcing of the pantograph in the video image,and the arcing time,times,arcing area,etc.are calculated,and the arcing detection performance of different methods is analyzed by comparing the results of the detection calculations.Experiments show that the feature fusion SVM can achieve better detection results for the pantograph arc detection,and the data set requirements are relatively low;the improved MobileNet-SSD algorithm can have higher detection accuracy while ensuring the detection speed;Compared with the MobileNet-SSD algorithm,the improved MobileNet-SSD algorithm and the feature fusion SVM algorithm have higher detection accuracy.Aiming at the detection and recognition of pantograph arcs,this thesis uses traditional algorithms and deep learning target detection and recognition technology to research and analyze.The research results have certain application value for future arc detection and recognition.