Lightweight YOLO Framework for Image-Based Railway Foreign Object Intrusion Intelligent Perception

Image detection plays a crucial role in preventing foreign object intrusion on railways. However, the high speed of trains requires high detection efficiency. And foreign objects must be detected from a long distance. As a result, foreign objects often occupy only a small portion of the image, leading to insufficient information. To address the challenges of slow foreign object intelligent perception, deployment difficulties, and low precision in small-object detection within railway scenarios, a lightweight YOLO framework based on YOLOv5s for railway foreign object detection(IRO-YOLO) is presented in this study. First, improvements are made to GhostNetV2 by introducing the GGCR module to reduce backbone computational complexity. Second, multi dimensional collaborative attention (MCA) is incorporated into the detection head to improve the focus on small foreign objects. Finally, Scale_EIOU is proposed to further optimize small-object detection during training. Experimental results demonstrate that, compared to the original YOLOv5, the proposed algorithm reduces parameters by 7.98%, FLOPs by 18.99%, and increases FPS by 29.54%. Furthermore, mAP50 and mAP50-95 improve by 2.41% and 10.28%, respectively. These results highlight the proposed model's significant capability in enhancing obstacle detection performance in railway environments.