路基探地雷达图像层位智能识别与状态评价  

Ground⁃penetrating Radar Image Layer Intelligent Recognition and State Evaluation of Subgrade

作　　者：杜翠 戴天 王科盛 张千里 刘杰 DU Cui;DAI Tian;WANG Kesheng;ZHANG Qianli;LIU Jie(State Key Laboratory of Coal Resources and Safe Mining,Beijing 100083,China;Railway Engineering Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;State Key Laboratory for Track Technology of High-Speed Railway,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;Wenzhou Railway and Urban Rail Transit Construction Management Center,Wenzhou Zhejiang 325000,China;Faculty of Engineering,The University of Hong Kong,Hong Kong 999077,China)

机构地区：[1]煤炭资源与安全开采国家重点实验室,北京100083 [2]中国铁道科学研究院集团有限公司铁道建筑研究所,北京100081 [3]中国铁道科学研究院集团有限公司高速铁路轨道技术国家重点实验室,北京100081 [4]温州市铁路与城市轨道交通建设管理中心,浙江温州325000 [5]香港大学工程学院,中国香港999077

出　　处：《铁道建筑》2023年第4期84-87,共4页Railway Engineering

基　　金：煤炭资源与安全开采国家重点实验室开放基金资助(SKLCRSM21KFA06);中国铁道科学研究院集团有限公司基金(2021YJ317)。

摘　　要：针对有砟铁路路基层位探地雷达图像人工追踪效率低、精度低的问题,提出了基于深度学习的路基层位智能识别与状态评价技术。根据多条线路现场实测数据建立样本集,基于YOLO v5和U-Net训练智能识别模型,提出了层位厚度、道床-基床界面平整度评价指标计算方法。结果表明:本文提出的有砟铁路路基层位智能识别方法具有较高精度,且U-Net模型识别效果优于YOLO v5模型,识别、评价结果与人工追踪结果偏差较小,满足铁路路基检测工程的实际需求。Aiming at the problems of low efficiency and low accuracy of manual tracking of ground‐penetrating radar image of subgrade layer of ballasted railway,an intelligent recognition and state evaluation technology of subgrade layer based on deep learning was proposed.Based on the on-site measurement data of multiple railways,a sample set was established,and an intelligent recognition model was trained based on YOLO v5 and U‐Net.A calculation method for evaluation indicators of layer thickness and subgrade interface smoothness was proposed.The results show that the intelligent recognition method for the layer of ballast railway subgrade proposed in this article has high accuracy,and the recognition effect of the U-Net model is better than that of the YOLO v5 model.The deviation between the recognition and evaluation results and the manual tracking results is relatively small,which meets the actual needs of railway subgrade inspection engineering.

关 键 词：有砟铁路 铁路路基 探地雷达 层位识别 图像分割 状态评价 

分 类 号：U213.1[交通运输工程—道路与铁道工程]