干旱区生态环境知识资源中心

Land cover classification is an important and basis work to desertification monitoring and control in the desertification area. In this paper, a hierarchical object oriented classification method for extracting the information of sandy land with ALOS Imagery is presented. The ample spectral features, textural features and customized features such as Vegetation Coverage are used synthetically. There are two steps in the hierarchical object oriented method. Firstly, the image is segmented with multi-resolution segmentation method. Secondly, the hierarchical classification rules are constructed, and the different features are used to distinguish sandy land and other typical land cover types in the nodes of the hierarchical classification rules, while sandy land is classified into mobile sand, semi-fixed sand and fixed sand. Through the confusion matrix, the total accuracy of the method proposed in this paper is enhanced from 69.73% (pixel-based minimum distance classification), 80.25% (object oriented nearest neighbor classification) to 85.19%. The result indicates that the hierarchical method with specific features gains a high classification accuracy and is with a high degree of automation.