Intrusion object recognition of rail perimeter using an improved YOLOv5 algorithm

Shiding Guo, Weibing Zhu, Yukai Feng, Lanai Cheng, Shaohua Miao, Lirong Lian, Tangwen Yang, Ye Li, Sitao Zhang, Lingyun Zhang, Xin Wang, Xin Yu

Abstract: To recognize intrusion objects in railway perimeter, a small object recognition algorithm is proposed here based on YOLOv5. An anchor frame layer for small object is added to the original YOLOv5 framework, and a SPD-Conv module is used in the YOLOv5 algorithm for performance improvement. The experimental results on the self-collected dataset show that the proposed algorithm can improve the average recognition accuracy of small object and effectively mitigates the missed detection and false recognition of small objects.