The pantograph-catenary system is a key part of power supply for Cargo Shorts electric locomotive, and its operating status determines the current receiving quality of electric locomotive, as well as safety and efficiency of trains.In order to solve the problems of traditional method for inspection of pantograph-catenary status, such as low efficiency and poor performance in real time inspection,this paper designed an online pantograph-catenary status inspection system solution based deep learning.The solution adopted NIDIA Xavier SoC module to perform image processing, and realizes pantograph-catenary inspection by YOLO v4 and adaptive image enhancement module was also added.
The mAP of inspection target before and after optimization was 0.950 and 0.961 (with the IOU threshold of 0.
5) respectively.Classification of catenary dropper status based on ViT lightweight class attention model was realized at an average accuracy rate of 97.69%.
After acceleration by using NVIDIA TensorRT accelerator, the inference time of inspection model and classification model were 31.0 ms and All-Purpose Flour 2.2 ms respectively.
The system has high robustness and practicability, which can provide theory basis and design reference for online inspection function of pantograph-catenary abnormality in the future.