高分辨率影像城区建筑物提取研究  

Urban building extraction from high-resolution multi-spectral image with object-oriented classification

作　　者：刘海飞[1] 常庆瑞[1] 李粉玲[1] 

机构地区：[1]西北农林科技大学资源环境学院,陕西杨凌712100

出　　处：《西北农林科技大学学报（自然科学版）》2013年第10期221-227,234,共8页Journal of Northwest A&F University(Natural Science Edition)

基　　金：国家自然科学基金项目(30872073);国家"973"重点基础研究发展计划项目(2007CB407203)

摘　　要：【目的】探讨高分辨率遥感影像城区建筑物提取方法,为快速获取城区建筑物分布和辅助制订城区发展规划提供参考。【方法】以陕西杨凌西北农林科技大学北校区为研究对象,采用知识规则与支持向量机(Support vector machines,SVM)相结合的面向对象分析方法,从QuickBird影像中提取建筑物,并与基于SVM的面向对象分析方法及传统的基于像元的分类方法进行比较。【结果】采用知识规则与SVM相结合的面向对象分析方法所得的分类结果表明,提取建筑物总体精度达到90.68%,Kappa系数为0.81,较基于SVM的面向对象分析方法、SVM、最大似然法、K均值法总体精度分别提高了10.38%,15.31%,26.4%和29.2%。【结论】基于知识规则和SVM相结合的面向对象分析方法精度高、速度快,可快速获取建筑物的分布情况。【Objective】This study explored the method of urban building extraction to help get building distributions faster and make better urban plan.【Method】The study extracted the building roof from QuickBird image of north campus of Northwest A＆F University using object-oriented classification with rules and support vector machine.The classification result was compared with object-oriented classification using SVM and traditional methods based on pixels.【Result】Classification results showed that the overall classification accuracy of object-oriented classification using rules and SVM was 90.68%,10.38%,15.31%,26.4%,and 29.2% higher than object-oriented classification using SVM,SVM classifier,maximum likelihood classifier and K-Means classier,respectively.【Conclusion】Objectoriented classification using rules and SVM led was an efficient way to get building distributions.

关 键 词：高分辨率 知识规则 支持向量机 多尺度分割 面向对象 精度评价 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]