Abstract: The catenary is a transmission line that provides power for the train, and foreign objects such as plastic bags attached to the catenary will cause potential safety hazards to the train operation. In order to solve the problems of low efficiency and high labor cost of manual inspection, a YOLO-CDBW model for catenary foreign body detection based on YOLOv7 is proposed. Firstly, in the feature extraction stage, a feature extraction module using residual bottleneck structure and depth separation convolutional layer is constructed to avoid the problem of small target feature loss caused by the increase of network depth and reduce the amount of network computation. Finally, the WIoU loss function is used to optimize the model and focus on the ordinary mass anchor frame through the dynamic focusing mechanism to improve the prediction accuracy. Experimental results show that, the average mAP_0.5 of the YOLO-CDBW model reaches 87.1% and the detection speed FPS reaches 66.5 frame/s, which are 5.0 and 10.8 percentage points higher than those of the YOLOv7 model, respectively, meeting the needs of catenary foreign body detection.

Key words: contact line, foreign substance examination, YOLO, object detection, loss function, attention mechanisms