铁道科学与工程学报2025，Vol.22Issue(9)：4204-4217,14.DOI:10.19713/j.cnki.43-1423/u.T20241899
动态感知与特征增强的轨道异物检测方法
Rail foreign object detection based on dynamic sensing and feature enhancement
摘要
Abstract
The problem of foreign object intrusion on railway tracks represents a core safety challenge for modern high-speed rail transportation systems.With the rapid development of transport infrastructure and its ongoing intelligent transformation,high-speed railway operations have imposed unprecedented stringent requirements on detection technologies.However,existing foreign object detection methods often struggle to achieve an optimal balance between detection accuracy and processing speed,making them inadequate for practical track monitoring applications.To address these challenges,this paper proposed a railway foreign object intelligent boost(RF-IB)method based on dynamic perception and feature enhancement.First,the fixed masking was employ to define region of interest(ROI)areas,focusing on critical zones along the track and its immediate surroundings.Second,a split reverse bottleneck(S-Bneck)module was constructed with a multi-branch processing strategy to overcome gradient attenuation during information transmission.Concurrently,a feature reuse module(FRM)was implemented to establish hierarchical recursive connections between feature maps of different scales,thereby enhancing the reusability of both semantic and detailed information.Additionally,an adaptive upsampling module(AUM)was designed,which employed a more efficient content-aware mechanism to restore detailed features while reducing computational overhead.Finally,an anchor-free detection network was introduced to improve real-time detection capabilities.Experimental results demonstrate that our method achieves mean average precision(mAP)scores of 57.0%and 47.3%on a custom railway foreign object dataset and the public COCO dataset respectively.With a compact model size of 35.2 MB and processing speed of 89.8 frames per second,the proposed approach effectively can meet the dual requirements of detection accuracy and real-time performance in railway track scenarios.关键词
轨道异物检测/动态感知/特征增强/反向瓶颈/无锚框检测网络Key words
track foreign object detection/dynamic perception/feature reuse/reverse bottleneck/anchorless box detection network分类
信息技术与安全科学引用本文复制引用
沈瑜,李博昊..动态感知与特征增强的轨道异物检测方法[J].铁道科学与工程学报,2025,22(9):4204-4217,14.基金项目
国家自然科学基金资助项目(61562057,61861025,62241106) （61562057,61861025,62241106）
智能化隧道监理机器人研究项目(中铁科研院(科研)字2020-KJ016-Z016-A2) （中铁科研院(科研）
四电BIM工程与智能应用铁路行业重点实验室开放项目(BIMKF-2021-04) （BIMKF-2021-04）
