Research on Small Target Railway Foreign Object Intrusion Detection Based on YOLOX- Tiny

Based on the YOLOX- Tiny detection model, a multi-scale fusion and attention-enhanced anchor-less deep learning model for foreign object detection was proposed, which solved the problems of insufficient feature extraction, low detection accuracy, and low real-time performance in railway foreign object intrusion detection. Based on the YOLOX- Tiny model multi-scale FPN pyramid network, a small target detection layer is added to propose a multi-scale weighted bidirectional feature fusion module, which enhances the utilization of feature maps and improves detection accuracy. At the same time, an index attention mechanism is introduced to improve the network structure of YOLOX- Tiny, increase the focus on target regions, and highlight target features in small target scenarios. This approach effectively improved the detection accuracy for small-scale targets and enhanced the convergence speed.