关键词: 农作物精细分类, Sentienl-1, Sentinel-2, 卷积神经网络, 雷达植被指数

Abstract: Optical imagery commonly used in crop classification is susceptible to interference from clouds and rainfall，which limits the application of remote sensing technology in agricultural monitoring for certain regions. Synthetic aperture radar（SAR）data has the advantage of being less susceptible to weather conditions.To investigate whether SAR images can be used to achieve accurate crop classification，four classifiers，namely convolutional neural network（CNN），extreme gradient boosting（XGBoost），random forest（RF），and support vector machine（SVM），were used to classify crops in the middle reaches of the Heihe River based on Sentinel‐1 backscatter coefficients and the dual polarimetric SAR vegetation index（SVI_DP）.The classification results were compared with those of Sentinel‐2 optical images.The results showed that the overall accuracies of the four classifiers（CNN，XGBoost，RF，SVM）were 81.50%，78.49%，77.92%，and 76.60% when using SAR images containing SVI_DP as training data，and 82.21%，79.23%，77.96%，and 76.34% when using optical images as training data，which were similar in classification accuracy.For complex categories such as alfalfa and others with intricate feature information，using SAR images could achieve higher accuracy.In conclusion，the radar vegetation index can enrich the feature information of SAR images，and SAR images can be applied to crop classification tasks and yield accurate classification results.

Key words: Fine crop classification, Sentinel‐1, Sentinel‐2, Convolutional neural network, Radar vegetation index