ENVI遥感图像处理中几种监督分类方法的比较  

The Comparison of Several Kinds of Supervised Classification Methods in ENVI Remote Sensing Image Processing

作　　者：金杰[1] 朱海岩[1] 李子潇[1] 孙建伟[2] 

机构地区：[1]河北工程大学水电学院,河北邯郸056021 [2]邯郸市水利局,河北邯郸056000

出　　处：《水利科技与经济》2014年第1期146-148,160,共4页Water Conservancy Science and Technology and Economy

基　　金：水利部公益性基金项目(漳河上游降雨径流形成及演变规律研究编号201201091)

摘　　要：遥感图像通过像元值的差异反映地物的光谱信息及空间变化原理来表示地物差异[1],为提取各类信息,图像分类尤为重要[2]。在遥感图像数据分类处理过程中,监督分类是普遍适用的一种分类方法[3]。为了评价几种监督分类的优势与缺点,主要选用SPOT5图像数据进行试验,介绍了监督分类中平行六面体、最小距离以及最大似然3种分类方法[4]。通过定义训练样本、训练样本评价、执行监督分类、分类结果评价等步骤[5],得出最大似然分类法的总体分类精度最高,制图精度和用户精度都很高,漏分误差和错分误差较低,而且计算时间较短,相对于另外两种分类方法更有应用优势。Through remote sensing image pixel value differences reflect the spectral band information and spatial variation principle to represent the feature differences between various types of information for the extraction, the image classification is particularly important. In remote sensing image process- ing, supervised classification is generally applicable to a classification method. In order to evaluate the advantages and drawbacks of several supervised classification, the paper chose the SPOT5 image data to test supervised classification and introduced parallelepiped method, minimum distance and maximum likelihood method. By defining the training sample, training sample evaluation, implemen- tation monitoring classification and classification evaluation, it shows that draw maximum likelihood classification has highest overall classification accuracy, precision and user mapping accuracy, the leakage of error and misclassification error is low, and the computation time is shorter, it has more application advantages than other two kind of classification method.

关 键 词：遥感 监督分类 平行六面体法 最小距离法 最大似然法 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]