铁道标准设计2025，Vol.69Issue(3)：48-57,10.DOI:10.13238/j.issn.1004-2954.202406190003
面向铁路设计控制要素智能解译的遥感影像样本库构建方法
Method for Constructing Remote Sensing Image Sample Database for Intelligent Interpretation of Railway Design Control Elements
摘要
Abstract
Automatically interpreting railway design control elements from remote sensing images is crucial for achieving"one-click mapping."However,deep learning-based intelligent interpretation of remote sensing images requires a large number of labeled samples.This paper proposes a method for constructing a sample database for the intelligent interpretation of design control elements using multi-source remote sensing data,based on the principles of railway alignment design.First,initial samples were automatically generated using multi-source data from Digital Orthophoto Maps(DOM),Digital Line Graphic(DLG),and Light Detection and Ranging(Lidar)point clouds.Next,an incremental active learning iterative approach was applied to optimize the initial samples to achieve high quality and comprehensive coverage of the railway lines.Taking the Changsha-Ganzhou Railway as an example,a high-resolution intelligent interpretation sample database was constructed,focusing on four types of design control elements(houses,roads,water bodies,and vegetation)along the railway line.This database,named the Wuhan University Sample Database of Control Elements of Railway Route Design(WHU-RRDSD),had a ground resolution of 0.1 m and a total of over 200,000 samples.Finally,to validate the usability of the sample database,a detailed validation was conducted from three aspects:qualitative evaluation,quantitative evaluation,and application in other scenarios.The results showed that the IoU evaluation metrics for the four types—houses,roads,water bodies,and vegetation—were 84.43%,82.38%,90.19%,and 90.28%,respectively,demonstrating excellent interpretation performance.The intelligent model trained on WHU-RRDSD was transferred to the interpretation of houses,roads,water bodies,and vegetation in the Yichang-Fuling high-speed railway scenario,validating the applicability of sample database in other scenarios.Additionally,two application cases based on the WHU-RRDSD sample—weakly supervised building extraction and object classification using high-resolution remote sensing images—were briefly introduced,further validating the usability of the sample database constructed by this method.关键词
长赣铁路/控制要素/样本库构建/深度学习/增量主动学习/影像解译Key words
Changsha-Ganzhou Railway/control elements/sample database construction/deep learning/incremental active learning/image interpretation分类
交通运输引用本文复制引用
冯海霞,胡庆武,王恬妮,柳天成,郑道远,曹成度..面向铁路设计控制要素智能解译的遥感影像样本库构建方法[J].铁道标准设计,2025,69(3):48-57,10.基金项目
国家重点研发计划项目(2021YFB2600401) （2021YFB2600401）
