基于知识的遥感图像分类方法的探讨

本文提出了一种基于知识的遥感图像土地利用分类法。包括知识的获取和表达 ,以及将知识用于土地利用分类。知识的获取和表达包括两方面。一是将原有的GIS土地利用数据与新的遥感图像配准、叠加、集成 ,并从中挖掘各类型的光谱知识。二是通过对研究区各土地利用类型变化特点进行分析研究 ,并从中得到原有各土地利用类型经数年后变成其它类型的可能性及其大小。并将其定量化表达成新遥感图像上各类型的先验性概率。在知识的应用和分类方面 ,分别对原土地利用各类型所对应的遥感图像进行知识的应用和处理 ,利用各自可能出现的类型及其各类型的先验性概率进行最大似然法分类。研究表明 ,该方法将原有的GIS数据和各种土地利用类型变化的先验性知识综合集成用于新的遥感图像的分类中。这不仅可以促进GIS数据更新的自动化 ,而且还可以得到比常规最大似然法高的分类精度。

Investigation on Classification of Remote Sensing Image on Basis of Knowledge

A new classification scheme, which includes acquirement and expression of knowledge, as well as application of knowledge in land use classification, is presented. The acquirement and expression of knowledge include two aspects: the first is that the spectrum knowledge of each land use type is automatically obtained from old land use map matched and overlain with remote sensing images; the other is that the changing knowledge of land use types are collected and analyzed, and changed into prior knowledge and probability. In application of the knowledge, maximum likelihood classification is respectively applied to every part of remote sensing images covered by individual land use type with its prior knowledge and probability.