论文部分内容阅读
1 引言在遥感影象数据分析中,最重要的任务之一就是把影象数据分成不同的地物类型,如土壤、水体和岩石等。在与政策有关的情况下,可能还需要诸如淹没面积,某种作物面积或海岸线长度这样一些准确的定量信息。这种准确的定量信息只有通过使用模式识别(PR)法则才能获得(Tou和Gonzales 1974)。实际上,影象分类是在纯场地和纯象元基础两种情况下实现的。纯场地分类,是将一个小的均一地区(一组彼此相邻的象元)看作所有象元均具有某种状况来进行分类。这类分类器最适于农田和林地的分类,因为就农田和林地的开阔性来说均一区
1 Introduction One of the most important tasks in remote sensing image data analysis is to divide the image data into different types of features such as soil, water and rock. In policy-related situations, accurate quantitative information such as submerged area, crop area, or coastline length may also be required. This accurate quantitative information is only available through the use of pattern recognition (PR) laws (Tou and Gonzales 1974). In fact, image classification is achieved in both pure-site and pure-image-based cases. Pure-field classification is to classify a small homogeneous area (a group of pixels that are adjacent to each other) as if all the pixels have a certain condition. This classifier is best suited for the classification of farmland and woodland because of the homogeneity of the openness of farmland and woodlands