A YOLO-Based Algorithm for Detecting Key Components on Subway Train Roof

The operational status of key components on the roof of subway trains is closely related to the safety of vehicle operation. In response to the practical need for detecting these roof components, this paper proposes a subway train roof key component detection algorithm based on YOLOv5s. First, considering the limited computational resources in the field, a lightweight design for YOLOv5s was implemented by replacing the C3 module in the YOLOv5s backbone with the Ghost_C3 module, and replacing the CBS module in the Neck layer with Ghost_CBS. Additionally, to compensate for the performance degradation caused by the lightweight design, a CA attention mechanism was introduced in the Ghost_C3 module to enhance the network' s ability to perceive the features of key components, thus improving the network' s accuracy. Experimental results showed that the improved YOLOv5s network model achieves a 1.36 % increase in mAP, a 33.98 % reduction in Params, and a 36.65 % reduction in FLOPs compared to the original YOLOv5s network, with a noticeable improvement in the AP metric for detecting key components of the train roof.