Abstract: Taking Shihezi county as study area, the best temporal combination of land use-cover remote sensing classification was presented in the paper by NDVI time series analysis in order to further improve classification accuracy in arid and semi-arid areas. And the optimum segmental bands were selected with OIF index method. Four segmental levels were selected based on different segmentation parameters and different scales classification objects were constructed. According to the characteristic of different kinds of objects, the process of object-oriented land use-cover classification was worked out by means of fuzzy classification knowledge-based and unsupervised classification example-based respectively. Classification effects were evaluated through ground test data of the study area. Comparing with the method of pixel classification, the results show that the new method has higher accuracy, and can be a reference for later researches and applications.