面向对象的玛纳斯河流域TM影像分类技术研究

Research for Object-Oriented TM Image Segmentation and Information Extraction Technology

以面向对象的遥感影像分类的理论与方法,基于ENVI 5.0、eCongition 8.7和ArcGIS 10.0平台,从遥感尺度问题和决策树分类技术入手,对面向对象分类方法的多尺度分割和多特征分类等关键问题进行了探讨,并应用该方法对玛纳斯河流域的信息进行了提取研究。结果表明,多尺度分割最佳波段权重为0,1,2,2,1,0,1,形状参数和紧致度均为0.7,分割尺度为60,30和10,地物种类可分为林地、草地、水域、耕地、人工表面、冰雪地、盐碱地、其他八大类。分类结果总精度为0.92,Kappa系数为0.91,证明面向对象的分类方法可以有效地对干旱区中尺度分辨率影像进行快速高效的地物信息识别及提取。

Based on the theory of object-oriented remote sensing classification and ENV15.0 eCongition8.7 and ArcGIS10.0,we discussed key issues on multiscale segmentation and feature extraction,and extracted the feature information of the Manas River Basin.Results show that the best multiscale segmentation band weight were 0,1,2,2,1,0,1,both shape parameter and firmness was 0.7,segmentation scales were 60,30 and 10.Feature types can be divided into woodland,grassland,water body,farmland,artificial surface,snow-coved land,saline-alkali soil and the other.The total accuracy of the classification results was 0.92 and Kappa was 0.91.It has been shown that the object-oriented classification method can effectively identify and extract the feature information of scale resolution images in the arid region.