Research On The Detection And Recognition System Of Pantograph End Horn Abnormality Based On Computer Vision

With the continuous expansion of the scale of high-speed railways in our country,railway transportation has become the main mode of transportation.The vigorous development of railway transportation not only saves a lot of resources but also provides many conveniences.The pantograph is located on the roof of the train and is the most important part of the train’s power supply system.The pantograph end horns ensure that the pantograph can safely pass through the switch of the catenary wire,prevent the pantograph from drilling into the switch and cause a safety accident.At the same time,it can contact the power supply line under extreme conditions to ensure the normal operation of the train.Traditional detection methods are inefficient,this paper used a method for detecting and classifying pantograph horns based on deep learning algorithms,it performs online detection and offline classification and storage of the abnormal conditions of pantograph end horns.The paper introduced the structure of the pantograph and selected the camera model,then expanded and enriched the dataset.In the image preprocessing,the classification of image illumination anomalies based on grayscale characteristics was used,and the fusion algorithm based on Gamma transform and CLAHE algorithm was used to correct the image for overexposed and underexposed images.Comparing each filtering method,taking PSNR as the filtering effect evaluation index,the effect of bilateral filtering is the best.The YOLOv4 and Faster R-CNN algorithms were used in the online detection of pantograph end horns to detect the abnormal state of pantograph horns.After data enhancement,labelimg was used to label the image data set,and the two algorithms were compared respectively.The results show that the deep learning method has high efficiency and accuracy,and has a significant effect in the online detection of pantograph end horns.The transfer learning method overcame the feature of deep learning relying on huge data,and carried out the offline classification and storage of pantograph horns.Considering that the relative positions of the pantograph and the detection camera are almost unchanged in engineering practice,the pictures of the pantograph end horns area were intercepted,classified and stored.The algorithms used include Res Net,VGG and Mobile Net networks,which were verified by actual experiments.The results show that these networks can have excellent performance in pantograph end horns classification,among which the Mobile Net network has the best performance.In order to improve the use efficiency better,this paper developed a pantograph end horns anomaly detection and classification system,integrated the algorithm of this paper into the system,and realized the more convenient use of the system.