基于指数计算的多层次遥感信息提取模型  

Index Based Hierarchical Model for Thematic Information Extraction

作　　者：郜丽静[1,2] 骆剑承[1] 沈占锋[1] 刘雯[1,3] 陈生[1] 

机构地区：[1]中国科学院遥感应用研究所,北京100101 [2]中国科学院研究生院,北京100080 [3]北京理工大学信息科学技术学院,北京100081

出　　处：《地理与地理信息科学》2009年第2期39-41,45,F0002,共5页Geography and Geo-Information Science

基　　金：国家863计划项目(2007AA12Z141、2007AA12Z224);国家自然科学基金项目(40601057、40871203)

摘　　要：指数计算是描述地物波谱特征简单而有效的方法,但单纯依靠指数方法进行遥感专题信息提取,错分、漏分现象严重。该文提出一种"全域-局部"的遥感信息提取思路:对整幅图像进行全域分割、全域分类,实现目标地物与背景地物的初始分离;在目标地物周围建立一定尺度的缓冲区,作为局部分割-局部分类的作用区域,并在局部作用域内选择不同的阈值,通过迭代过程实现目标地物与背景信息的最优分离。选取鄱阳湖地区Landsat 7/ETM+影像,应用该方法进行水体提取试验,精度达95.751%,优于常规的最大似然法。Index computation is a simple and effective way to describe features＇ spectral characteristics,but not very precise.This paper proposes a new method called＂Whole-to-Local＂iterative information extraction model.It is implemented at two different levels：global segmentation and classification,and local segmentation and classification.Global segmentation and classification were carried out,and target objects were mainly separated from background;then a buffer around the objects was set up as the local process areas,local-level processing was then implemented iteratively with different-level knowledge fused in to achieve the optimum edge and gain better accuracy in information extraction.In the paper,Landsat ETM＋ data of Poyang Lake was taken as an example to explain how the method worked in water-body extraction.The results were compared with traditional methods,and the precision reached to about 96%.

关 键 词：遥感信息提取 指数计算 分层分类 全域-局部 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]