GE Jia, ZHANG Ziming, WU Cheng, ZHAN Qian, SUN Yongjun
In this paper, part of the loess plateau was chosen as the study area. A set of automatic information extraction methods for water-erosion desertification was proposed by using the ETM+ images obtained in this area and on the basis of remote sensing data and geographic information system. NDVI (normalized difference vegetation index), KT3 (KT transform, humidity), slope, DEM (elevation) and typical feature spectral data were used to establish the characteristic bands of the study area,and then a decision tree classification rule could be constructed,which could exclude the non-water erosion desertification information effectively in the study area. The object-oriented muti-scale segmentation technology was adopted, and the slope, gully density and vegetation coverage were taken as the characteristic bands of the water-erosion desertification classification. With the building of the multi-characters space, the weight value was determined by the analytic hierarchy process, which also served as the classification index of the water-erosion desertification. The consistency of the evaluation between the automatic extraction results and the visual interpretation results shows a good linear relationship, with the overall consistency reaching 82.8%.