高分遥感影像不同形状建筑物半自动提取与规则化  

Semi-automatic Extraction and Regularization of Buildings of Different Shapes from High-Resolution Remote Sensing Images

作　　者：崔卫红[1] 李佳 刘宇[2] CUI Weihong;LI Jia;LIU Yu(School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,Hubei,China;CETC Key Laboratory of Aerospace Information Applications,Shijiazhuang 050081,Hebei,China)

机构地区：[1]武汉大学遥感信息工程学院,湖北武汉430079 [2]中国电子科技集团公司航天信息应用技术重点实验室,河北石家庄050081

出　　处：《应用科学学报》2022年第3期372-388,共17页Journal of Applied Sciences

基　　金：国家自然科学基金(No.U2033216)资助

摘　　要：现有的高空间分辨率遥感影像交互式建筑物提取方法需要用户在建筑物上勾画出与建筑物大小和形状相近的线,且大多方法只能提取直角建筑物。为降低交互要求并实现不同形状建筑物的精确提取,该文首先在用户少量交互的基础上采用区域生长、高斯混合模型、CannyLines线段检测算法以及基于多星形约束的最大流/最小割分割模型获得建筑物图斑,然后分别针对直角建筑物和非直角建筑物图斑进行规则化,得到与实际建筑物形状一致的提取结果。实验表明,该方法交互简单且建筑物提取精度F1值可达到0.9,具有较强的鲁棒性。The current methods of interactive extraction of buildings from high-resolution remote sensing images mostly require complex user interaction and most of them only support extraction of buildings with right angles.In order to reduce interaction and achieve high-precision extraction of buildings in different shapes,this paper uses region grow,Gaussian mixture models(GMM),CannyLines edge detection and the max-flow/min-cut segmentation method based on multiple star constraints sequentially to obtain building patch,followed by regularization methods to get the building contours which are consistent with the actual building shapes.The average of F1is up to 0.9 in extraction experiments,and the experimental results also show the facility and strong robustness of the proposed method.

关 键 词：高分遥感影像 半自动 建筑物提取 规则化 

分 类 号：P237[天文地球—摄影测量与遥感]