A Study on Foreign Object Intrusion Detection for Trams Based on the STA-YOLO

Compared with trains and subways, the operating environment of trams is more open and complex, and the types of track foreign objects are also more diverse, and the existing research mainly focuses on closed scenarios, and there is a relative lack of research on the complex environment of trams. Therefore, this paper proposes the STA-YOLO algorithm with improved YOLOv11n to enhance the detection performance in the complex traffic environment of trams. The algorithm introduces the Scale Sequence Fusion Module (ScalSeq) to enhance the multi-scale modelling capability, adopts the Triple Feature Encoder (TFE) to improve the recognition effect of dense overlapping targets, and fuses the multi-scale and detailed features through the Channel and Position Attention Mechanism (CPAM), and finally introduces the CBAM Attention Mechanism in the three key output branches to strengthen the attention to the channel and spatial key information, to accelerate model convergence. The proposed algorithm was experimentally evaluated on both a self-built dataset and a public railway foreign object dataset, achieving a maximum [email protected] of 87.2% and an inference speed of 162 FPS. It significantly outperformed models such as YOLOv11n and SD-YOLO. The results demonstrate that the algorithm offers advantages in detection accuracy, inference speed, and model complexity, effectively improving detection precision while ensuring real-time performance.