关键词: 荒漠化土地, 自动识别, 决策树, 分层提取, 研究探讨

Abstract: This paper is given for studying and discussing the auto-classification of desertification land types by using Landsat TM data in a typical desertification region of the oasis of Minqin county,Gansu province.Based on the desertification land classification system,the structure of the decision tree and the locations for each land-covered in the tree are determined.According to the spectrum-reflecting characteristic of targets and the integrated features of TM image data,identifying and extracting methods are correspondingly used in order that the remote sensing data are drawn on maximally.At first,the non desertification lands can be extracted from original TM image by using characteristics of Soil-Adjusted Vegetation Index(SAVI) and Digital Numbers(DNs).Then the major research focuses on the multi-layer information discernment and extraction of three types of desertification land-covers and their respondent grades.Each grade is an object and is processed to yield a layer.Overlaying every layer can create a complete map of desertification land types.In the image processing,the spectrum reflecting properties,statistics training sites Supervised Classification(SC),geometry and texture properties analysis and Normalized Difference Snow Index(NDSI) have been respectively utilized for distinguishing and classifying the different land types.The results show the application of decision tree and multi-layer technology could decrease the possibility of interaction and impact with others message and make target simplified in pixel identification.Meanwhile,the investigation and adjustment in the fields is quite important for analyzing the land-cover distribution and comparing the difference between grades.

Key words: desertification land, auto-identification, decision tree, multi-layer extraction technology, primary study and discussion