YOLO-Based Third-Rail Defect Detection With BGAN-Augmented Training

The third rail is a key component for powering the train, but it may be damaged by vibration during operation, affecting the stability of power transmission. To solve this problem, this paper proposes a third rail defect detection method based on YOLO and camera sensing. This method is beneficial to camera detection and reduces labor and time costs. Since defect image samples are limited, this paper uses GAN to generate additional defect images to expand the training data, reducing the risk of overfitting due to insufficient data. Experimental results show that this method can effectively improve the detection performance of each YOLO version and alleviate the data imbalance problem. The results show that YOLOv8n and YOLOv8s achieve mAP95 of 0.935 and 0.975 respectively. YOLOv9-tiny achieves a mAP95 of 0.99, while YOLOv9s achieves 0.994. In YOLOv10n and YOLOv10s, 0.935 and 0.942 are obtained. Overall, it is proposed to effectively improve the detection accuracy of YOLO by enhancing the dataset through GAN.