Small-Target Infrastructure Detection with Spatial Tracking and Multi-Level Feature Extraction of Railway Track Scenes

In the field of railway transportation, with the continuous improvement of train running speed and the continuous increase of operating mileage, it becomes particularly important to ensure the safety and integrity of small-target infrastructure along the railway train. Although many small-target detection methods have been proposed, the research on small-target infrastructure detection along rail trains is very rare. To solve this problem, this paper proposes a small-target infrastructure detection network with spatial tracking and multi-level feature extraction based on Yolov8n. The experimental results show that ST-YOLO has a 3.6% mAP improvement on the OSDaR23 dataset compared to the baseline Yolov8n, and shows significant advantages over other state-of-the-art methods.