干旱区生态环境知识资源中心

Whether the selection of remote sensing features can he performed effectively will directly affect the quality of classification results. This paper proposes a multi-featured modeling strategy for the desert/grassland biome transition zone to improve the classification accuracy, taking Luoshan area of Ningxia province in China as the studying case and Landsat 8 OLI images of three periods as data source. The sparse representation based on dictionary learning is used as a classifier in order to select the combination of optimal features according to the multi-featured modeling strategy. It shows that the combination of spectrum, vegetation, terrain, architecture and water information can effectively improve the classification accuracy and reduce the classification uncertainty in the desert/grassland biome transition zone, and the best identified feature combinations includes b1 similar to b7, NDVI, DEM, NDBI, VAR(b5), MNDWI. Then, the statistical analysis of land cover changes in the study area from 2013 to 2015 was conducted.