基于面向对象和规则的遥感影像分类研究  

Classification of Remot Sensing Image Based on Object Oriented and Class Rules

作　　者：陈云浩[1] 冯通[2] 史培军[1] 王今飞[3] 

机构地区：[1]北京师范大学资源学院 [2]香港中文大学地理与资源管理系 [3]加拿大西安大略大学地理系

出　　处：《武汉大学学报（信息科学版）》2006年第4期316-320,共5页Geomatics and Information Science of Wuhan University

基　　金：RGCGrant资助项目(CUHK.4251/03H);国家自然科学基金资助项目(40201036);测绘遥感信息工程国家重点实验室开放研究基金资助项目(WKL(03)0102)

摘　　要：讨论了面向对象和规则的光学遥感影像分类方法。首先利用多尺度分割形成影像对象,建立对象的层次结构,计算对象的光谱特征、几何特征、拓扑特征等,利用对象、特征形成分类规则,并通过不同对象层间信息的传递和合并实现对影像的分类。并以北京城市土地利用分类为例,对该方法进行了验证。A classification method for remote sensing image based on object oriented and class rules is presented in this paper. The whole image is firstly segmented into image objects and object hierarchy is built using multi-resolution segmentation. Then, the features of objects, such as spectral, geometrical and topological characteristics, are measured. The class rules are constructed based on these features. Classification of remote sensing image is carried out using the rules and information transfering and merging among these objects. The accuracy of the classification approach based on object oriented and class rules is checked by a case study of land use classification of Beijing.

关 键 词：面向对象 多尺度分割 规则 分类 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置] P237.4[自动化与计算机技术—控制科学与工程]