【目的】旨在准确获取苜蓿（Medicago sativa L.）生产空间分布信息，为管理部门提供饲草饲料安全供给情况和技术支持。【方法】本研究使用Google Earth Engine(GEE)平台获取和处理哨兵（Sentine）系列卫星Sentinel-1和Sentinel-2数据，构建了光谱特征、植被指数和雷达极化特征数据集。利用决策树、随机森林、支持向量机和深度学习四种分类算法，评价不同特征组合遥感分类精度，筛选出最优的特征组合和苜蓿遥感识别模型。【结果】结果表明，结合光谱特征、植被指数和雷达极化特征的深度神经网络模型在苜蓿遥感分类中表现最优，总体精度为94.85%,Kappa系数为94.2%；研究证明了机器学习方法在提高苜蓿遥感分类精度方面的有效性；在四种分类模型中，光学特征加雷达特征综合组合的精度均为最高，说明了多源遥感信息对于提高模型性能的重要性。【结论】基于深度神经网络的苜蓿遥感识别模型，在结合光谱特征、植被指数和雷达极化特征的特征组合条件下，识别效果最好。

【Objective】In order to accurately get information on the spatial distribution of alfalfa,so as to provide security supply status of fodder grasses as well as technical support for related authorities.【Methods】Datasets of spectral features,vegetation indices and radar polarization features were constructed by Sentinel-1 and Sentinel-2 data of Sentinel satellites that were acquired and processed from the Google Earth Engine platform. Then four classification algorithms consisting of decision tree,random forest,support vector machine and deep learning,were utilized to evaluate the remote sensing classification accuracy of different feature combinations,so as to screen out the optimal feature combinations and remote sensing classification model of alfalfa.【Results】The results showed that:The deep neural network model combining spectral features,vegetation indices and radar polarization features performed optimally in alfalfa remote sensing classification,with an overall accuracy of 94.85% and a Kappa coefficient of 94.2%.The study demonstrated the effectiveness of machine learning methods in improving the accuracy of remote sensing classification of alfalfa.The combination of spectral features and radar features had the highest accuracy for all four classification models,indicating the importance of multi-source remote sensing information for improving model performance. 【Conclusion】 The deep neural network-based remote sensing identification model for alfalfa was best under the condition of combining the spectral features,vegetation indices and radar polarization features.