Abstract: With the continuous exploitation of mineral resources,land use types in mining areas change frequently. Obtaining the results of long-term land use classification in mining areas is of great significance to land management and ecological reconstruction in mining areas. Based on the Google Earth Engine,this paper constructs a multidimensional feature set for sentinel-2 and Landsat data from three aspects:spectral features,texture features,and terrain features. The random forest algorithm was used to build different feature models separately and test the accuracy to filter out the optimal classification model for the mining areas. A long-time series land use classification study was conducted by using the open-pit mine in Xilinhot as the study area in recent 30 years (from 1991 to 2020). The results indicate that:① Based on the GEE remote sensing cloud platform, the land use classification results of the study area in the past 30 years from 1991 to 2020 can be extracted efficiently,quickly and accurately. ② The spectral features play a decisive role in classification accuracy,and the incorporation of texture features and terrain features can effectively improve the accuracy of the land use classification model in mining areas. ③ The unique red-edge band of Sentinel-2 has a high sensitivity to vegetation,which can effectively improve the classification accuracy. The long-term continuous monitoring results can effectively understand the changes and laws of land use in Xilinhot open-pit mining area,and provide a solid data support for further analysis of the impact of human production and living and environmental changes on land use.