Research on Pantograph Condition Monitoring Based on Video Image

During the high-speed operation of the train, sparks are generated between the pantograph and the contact line, and electrical wear affects the reliability of the train’s pantograph network, which may cause operational safety accidents in severe cases. In this paper, a video image-based pantograph spark recognition algorithm is proposed. The pantograph area can be determined through the YOLOv4 neural network algorithm. The RGB and HIS color model criteria are used to obtain the spark color feature area, combined with the R component dynamic background differential judgment. The Wiener filter algorithm is used to process the dithered and blurred image. According to the spark generation area, the three-frame dynamic background subtraction and gray difference area criterion are used to eliminate the background interference of train operation, which can accurately identify the intensity of spark discharge in the video image, and the frequency of sparks can be combined to give an alarm. The simulation experiment results show that the detection accuracy of the spark recognition algorithm proposed in this paper is 98.0%, which significantly improves the accuracy of spark recognition and detection.