Cotton is one of the important economic crops in China, which is also a significant national strategic reserve. Timely and accurate acquisition of cotton spatial distribution is of great significance for ensuring the safety of the cotton industry. This paper takes Tumushuke City in southern Xinjiang, China as the research area, and uses Google Earth Engine (GEE) platform to obtain the data of Sentinel-2 and Landsat-9 from July to August in 2022, respectively. random forest (RF), support vector machine (SVM), classification and regression tree (CART) are used to extract cotton, and the best classifier is obtained, The texture features and optimal classification subset were used to obtain the optimal classification scheme, and the classification results of Sentinel-2 and Landsat-9 were analyzed and compared.. The results show that RF is the best classifier. The combination of spectral features, vegetation index and texture features has the highest classification accuracy. The overall accuracy (OA) of Sentinel-2 imagery is 92.1%, and the Kappa (k) is 0.88, while they are 89.3% and 0.85 for Landsat-9 imagery. And the classification accuracy of the selected features is close to the classification accuracy of the combination of spectral features, vegetation index and texture features. More specifically, the OA of Sentinel-2 imagery is 91.6%, and the k is 0.88, while they are 88.3% and 0.84 for Landsat-9 imagery. In general, the performance of Sentinel-2 is better than Landsat-9. On the one hand, the resolution of Sentinel-2 imagery is higher than Landsat-9. On the other hand, the red edge band of Sentienl-2 is beneficial to cotton extraction.