高分辨率多光谱影像城区建筑物提取研究.pdfVIP

39 6 Vol. 39, No. 6 2010 12 Acta Geoda etica et Ca rtographica Sinica Dec. , 2010 : 2010) 谭衢霖 , 100044 Urba n Bui lding Extra ction rom VHR Multispect ra l Images Using Obj ectbased Classi icat ion TAN Qul in Schoo l o Civil Engineering, Be ij ing Jia otong Unive rsity, Be ij ing 100044, China Abstract : Building extra ction i n urba n env i ronment requi res high spat ia l resolution remotely sensed da ta . However, tra ditiona l pixel based classi iers based on spectra l eatures a re i ne ective or highresol ution multi spectra l images due to la rge withinclass spectra l va riations a nd betweenclass spectra l con usions. In this study , a rulebased obj ectoriented classi ication method or bui ldi ng ext raction was developed rom a n IKONOS urba n scene. The method i ncluded the ollowi ng st eps: usion o 1 m pa nchromatic a nd 4 m multispectra l ba nds to produce a pa nsha rpened 1 m multi spectra l image; segment ation o the 1 m dat aset; ! supervised obj ectba sed classi ica tion into broad spectra l classes; ∀ spectra l, spatia l, text ura l a nd contextua l pa ra meters developed rom sa mple building objects were implemented in a uzzy logic rule ba se t o sepa ra te building roo tops rom other i mperv ious sur ace classes. The ruleba sed method identi ies bui ldi ng roo tops with 93 % accura cy in the experiment. Key words : obj ectoriented image a na lysis; uzzy logic; segmentation; classi ication; building ext raction : 城区高空间分辨率遥感数据由于存在大量同物异谱和异物同谱现象, 应用传 的基于像元光谱分类的方法进行 建筑物分类提取难以取得满意的效果发展一种从高分辨率IKONOS 卫星影像上基于知识规则的面向对象分类提取城 区建筑物方法, 包括如下步骤: 融合1 m 全色和4 m 多光谱波段影像, 生成1 m分辨率的多光谱融合影像; 分割融 合影像; ! 执行基于对象光谱的最近邻监督分类; ∀ 应用模糊逻辑分类器结合光谱空间纹理和上下文特征等知识规 则进行建筑物分类 计试验表明, 该分类方法提取城区建筑物精度可达到93 % : 面向对象影像分析; 模糊逻辑; 分割; 分类; 建筑物提取 : P237 : A : 国家自然科学基金( 4080 112 1) ; 中央高校基本科研业务费专项项目( 2009J BM067)