Subway Pantograph Arcing Detection Based on YOLOv10-CSEC

To address the challenges of short duration and low real-time detection accuracy of arcing between the pantograph and catenary in subway systems, this study presents a machine vision-based detection method utilizing the YOLOv 1 O-CSEC (CBAM, SE, ECA, CBAM) algorithm. The YOLOv10-CSEC algorithm improves arcing recognition accuracy by integrating SE, ECA, and CBAM attention mechanisms into the base YOLOv 10 algorithm, thereby enhancing both the channel and spatial dimension features. The proposed algorithm was evaluated using a practical arcing dataset sourced from the Nanjing subway system. The experimental findings indicate that YOLOv10-CSEC achieves a 3% improvement in detection accuracy and average accuracy improved by 2.8% compared to the standard YOLOv10 algorithm, while the detection speed was improved from 116 FPS to 125 FPS.