基于多尺度特征融合的轨旁异物侵限目标检测方法研究

打开文本图片集

中图分类号：U283.2 文献标志码：A文章编号：1006-0316（2025）12-0061-07

doi：10.3969/j.issn.1006-0316.2025.12.009

Abstract ∵ Railway foreign object intrusion detection technology is a crucial component ofrailway safety warning systems.However，existing detection methods based on machine vision are stillimited by accuracyand real-time performance.To enhance the performance of model detection，this study integrates StarNet into the C2f module and proposes the C2f-sn module.By improving the PAFPN based on YOLOv8，a multi-branch fused feature pyramid is introduced，where the CSP （Cross Stage Partial） concept isapplied to improve the Omni-Kernel， resulting in a CSPOmni-Kernel for enhanced feature fusion.Furthermore，a lightweight detection head structure based on shared convolution is designed，significantly reducing network complexity through aparameter-sharing mechanism. Finally， ablation and comparative experiments are conducted on a self-built multi-scenario railway foreign object dataset， with a focus on key metrics such as mean average precision （mAP50）and detection speed （FPS）.The experimental results demonstrate that the improved method achieves a 5.8% increase in mAP50 compared to the original model，and the detection speed improves from131.6 FPS to 237.3FPS，and the computational load decreases byO.2 GFLOPs.These findingsindicate that the proposed enhancements significantly boost detection performance，which provides a quantitative basis for practical applications.

Key words ∵ railwayforeign object intrusion ; object detection ; feature extraction ; shared convolution ; high-speed train

在我国铁路快速发展的背景下，轨道限界区域异物入侵问题已逐渐演变为影响铁路安全的主要问题。（剩余7856字）