一种高分辨率遥感影像道路提取方法  

A Road Extraction Method for High Resolution Remote Sensing Images

作　　者：戴激光[1,2] 朱婷婷 张依蕾 马榕辰 王晓桐 张腾达 DAI Ji-Guang;ZHU Ting-Ting;ZHANG Yi-Lei;MA Rong-Chen;WANG Xiao-Tong;ZHANG Teng-Da(School of Geomatics,Liaoning Technical University,Fuxin 123000;Beijing Key Laboratory of Urban Spatial Information Engineering,Beijing 100038)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,阜新123000 [2]城市空间信息工程北京市重点实验室,北京100038

出　　处：《自动化学报》2020年第11期2461-2471,共11页Acta Automatica Sinica

基　　金：国家自然科学基金(41871379);自然资源部国土卫星遥感应用重点实验室经费资助项目(KLSMNR−202004);辽宁省教育厅服务地方项目(LJ2019FL008);城市空间信息工程北京市重点实验室(2020221);地理国情监测国家测绘地理信息局重点实验室(2018NGCM01)资助。

摘　　要：针对空间异质性导致的道路几何纹理特征突出性下降问题,提出一种高分辨率遥感影像道路提取方法.首先设定跟踪模型,依据人工输入点,自适应提取道路中心点和道路宽度,设计迭代内插、双向迭代两种跟踪方式以及矩形跟踪模板;然后提出多描述子道路匹配模型,针对道路几何纹理特征突出性不足问题,基于道路区域地物边缘与道路方向一致的语义关系,通过线段峰值约束的思想,提出一种多尺度线段方向直方图(Multi-scale line segment orientation histogram,MSLSOH)描述子,以此对跟踪方向进行预测;针对道路几何纹理特征均质性下降问题,从道路区域与道路非道路混合区域纹理差异性出发,组合三角形构成扇形描述子,突出道路影像纹理特征,以此不仅可对预测跟踪点进行验证,而且也可在结构信息缺失的情况下对道路进行跟踪;最后选取不同类型、不同分辨率、不同场景的高分辨率遥感影像,通过与其他方法的实验对比,表明该方法能够解决道路提取过程中几何纹理特征突出性下降问题,具有准确率高和自动化程度高的优势.In order to solve the problem of road geometric texture feature prominence decline caused by spatial heterogeneity,a road extraction method for high resolution remote sensing images is proposed in this paper.This method first sets a tracking model,adaptively extracts road center point and road width according to manual input points,thereby designing two tracking methods(interpolation and bidirectional iteration),and a rectangular tracking template.Secondly,a multi-descriptor road matching model is proposed:Facing with the lack of prominent geometric features of roads,based on the semantic relationship between the edges of the objects in the road area and the direction of the road,a MSLSOH(Multi-scale line segment orientation histogram)descriptor is proposed to predict the matching tracking direction through the idea of line segment peak constraint.And then,for the problem of the homogeneity reduction of road texture features,from the difference of texture between road area and road nonroad mixed area,a sector descriptor composed of multiple triangles was designed to highlighting the road image texture features,so that not only the predicted tracking points can be verified,but also tracking roads in the absence of structural information.Finally,high resolution remote sensing images of different types,resolutions and scenes are selected.Compared with other methods,the experimental results show that this method can solve the problem of decreasing the prominence of geometric texture features in the process of road extraction,and has the advantages of high accuracy and high degree of automation.

关 键 词：语义关系 MSLSOH 描述子 扇形描述子 道路提取 高分辨率 遥感影像 

分 类 号：P237[天文地球—摄影测量与遥感] TP751[天文地球—测绘科学与技术]