An innovative application of pantograph recognition system based on deep learning

Abstract

One of the most significant aspects for the correct operation of a modern railway electrification system is the health of the pantograph and overhead line. The application of machine vision technology to monitor the status of pantographs in real time can reduce pantographs and catenary accidents caused by unpredictable events. It is challenging to achieve real-time and accurate criteria with present pantograph detecting technologies. Therefore, this methodology collects pantograph images through high-definition cameras and transmits them to the cloud through 5G, use the Mask R-CNN algorithm to process and analyse the images., This technology can assist railway technicians in judging the status of the pantograph. Mask R-CNN employs the Resnet network for feature extraction. Resnet has the characteristics of cross-layer connection, which avoids the problem of network degradation due to the deep learning network being too deep, and greatly improves the training efficiency. The recognition matching degree of pantographs are greater than 0.975, enabling pantograph recognition in a variety of environmental conditions. The use of 5G connection increases transmission speed and allows for real-time detection of pantograph status, which is critical for the railway's automated operation.