基于高分一号影像的土地覆被分类方法初探

Classification method of land cover based on GF-1 image

高分一号是我国发射的第一颗高分辨率卫星,其包含地物信息较为丰富,较多的应用于土地覆被分类中,但高分影像普遍存在基于像元分类精度稍低的问题,为了提高遥感影像的分类精度,基于高分一号影像,以新疆艾比湖湿地保护区为研究样区进行土地覆被分类研究。利用灰度共生矩阵方法提取图像的纹理信息,并将结果作为参数量输入到支持向量机(SVM)分类器中,将研究结果与传统的SVM分类及最大似然分类法作对比分析可得:辅以纹理特征的SVM分类方法可更好的区分地物信息,分类精度高达93.64%;传统的SVM分类精度为92.27%;最大似然分类为87.90%;因地制宜的开展辅以纹理特征的SVM分类方法是提高土地覆被监测精度的有效手段。

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 x 5 pixels, and the moving step as （ 1,1 ）. Comparing the result of classification to that of the traditional SVM classification and maximum likeli-hood 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.