Abstract: To further improve classification accuracy of low-resolution remote data in arid and semi-arid areas, taking Shihezi county in Xinjiang province as the study area, sample windows were selected by combining PSA (purposive selection algorithm) algorithm and statistical properties of region features distribution and finally the best sample window combinations were identified. Authentic membership function was obtained by probability density estimation. Then features identifying model of the region was constructed based on category membership function, and the remote classification flowing chart of large-scale land using/covering was established by using multi-resolution data. The result showed that the classification accuracy of low-resolution data were effectively improved by extracting exquisite distribution characteristics of features in various regions through the high spatial indentifying data, compared with method of Erdas unsupervised classification, the accuracy of fuzzy classification method was improved 20%. The research provides a useful reference and guidance in the researching and application of low-resolution data.