基于决策树规则的面向对象遥感影像分类  

Object-oriented Classification of Remote Sensing Image Based on Decision Trees

作　　者：葛春青[1] 张凌寒[2] 杨杰[1] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,武汉430079 [2]武汉大学遥感信息工程学院,武汉430079

出　　处：《遥感信息》2009年第2期86-90,56,共6页Remote Sensing Information

基　　金：高分辨率遥感影像三维测图与更新的理论和方法研究项目(40523005);山东省基础地理信息与数字化技术重点实验室开放基金资助项目

摘　　要：高分辨率影像包含了丰富的空间信息,这使得基于像素的传统分类方法的分类精度受到局限。基于此,本文对面向对象的分类方法进行了探讨。首先,利用图像的光谱和形状因子对图像进行合理的分割。然后,建立决策树分类判别知识库,将对象归属到某一类上并进行分类。结果显示,面向对象方法的分类精度较传统分类方法有了很大程度的提高,这为通过建立决策树知识库对地物光谱混杂的城区分类提供了一种有益的尝试。High resolution remote sensing image contains rich spatial information, for which pixel-based traditional classification methods can＇t satisfy the accuracy requirements. Accounting for this request, object-oriented image analysis method is presented in this paper. Firstly, the image is properly segmented using its spectral and shape factors. Then the distinguishing knowledge base of decision trees is built and objects are assigned to some classes. The results show that the object-oriented approach gives more accurate results than those achieved by traditional classification algorithms, and that it can provide a useful attempt to urban classification of hybrid object spectrum by building the distinguishing knowledge base for decision trees.

关 键 词：高分辨率影像 面向对象 多尺度分割 知识库 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]