Abstract: In recent years,a series of ecological problems such as water resource declining,watershed forest retreat, sand and saline increase at Ebinur Lake,Xinjiang wetland areas have become more and more serious due to the lower intensity agricultural production,population growth,irrational use of resources,and is threatening the sustainable development of the region directly. GF-1 was the first launch of a high-resolution satellite in China, and contained much ground object information which had been already applied in land cover classification. How to make better use of the GF-1 image information in extracting land cover types quickly has become an important topic in China's satellite application research. In order to meet the demand of land cover classification through remote monitoring in Ebinur Lake wetland area,this paper carried out a research of land cover classification by GF-1 image. The main method is utilizing GLCM to extract four textures information,and selecting the contrast ratio as the best texture through comparative analysis of the actual situation in the study area. Finally,enter it into the SVM classifier. Select the window for extracted texture features as 5 × 5 pixels,and the moving step as(1,1). Comparing the result of classification to that of the traditional SVM classification and maximum likelihood classification methods,it showed that the added texture information to SVM classification had high accuracy, which was up to 93.64%;traditional SVM classification accuracy was 92.27%;maximum likelihood classification accuracy was 87.90%. The innovation of this paper is it is the first time to add the texture to SVM classification and apply to GF-1 image. And it is very rare to apply this classification method to arid and semi-arid regions, which can be carried out to extract the internal oasis land use types effectively,and provide assistance for land-use planning in this area. However,the texture information input in this experiment was the contrast ratio,in fact,there are a lot of other texture features such as variance,mean,uniformity,and so on that can be studied in future. Texture features can be chosen based on the specific feature information of the experimental area. The method to extract texture information are various,using different methods to get the texture feature information for further analysis will be the focus of future research,which can provide reference for the GF-1 application in the land resources research and improving GF-1 satellites on land cover use in profit and monitoring capabilities.

Key words: GF-1, classification of land cover, texture information, SVM