Abstract: In order to solve the problems of small and broken parcels, complex planting structure, entanglement of crop remote sensing spectral features and difficulty in accurate information extraction in the hilly area of the middle and lower reaches of the Yangtze River, a more precise identification method of crop planting structure based on Sentinel-2A data is proposed in this paper. Firstly, the key phenological features of the main crops in the study area are obtained. Secondly, the spectral features, texture features and terrain features are calculated to construct the original feature sets. Finally, the importance of the features is sorted by using the random forest method, the feature variables of the original feature set are optimized, and the optimized combination features are selected for supervised classification to extract the crop information in the study area. The results show that:compared with the univariate feature, the overall classification accuracy and kappa coefficient are improved from 80.4% and 0.748 to 96.3% and 0.954 respectively which effectively improves the accuracy of crop classification in southern hilly area, and the algorithm robust is more stable.

Key words: crops, identification, Sentinel-2A, feature optimization, feature extraction