大数据驱动训练的深度学习模型是当今农作物分类的最新方法，有越来越多的新的分类模型涌现。然而，当前研究仍然主要关注模型方法的创新性，其应用方式往往针对特定时间、特定地区，而作物分类模型的泛化能力分析经常被忽略。提高遥感分类模型在大尺度空间范围的有效迁移能力是遥感技术支撑地球系统科学研究和社会应用的关键。本研究通过设计实验分析了模型架构、作物物候特征、农区地块尺度、数据类型等因素对作物分类模型泛化能力的影响。结果表明当训练区和测试区地块大小发生明显变化时，MultiResUNet相对于SegNet,　DeepLab V3+和U-Net具有更好的泛化性能。然而，单纯依靠MultiResUNet的泛化能力依然无法完全克服地块空间形态的变化对模型迁移的不利影响，为获得更高精度的华北玉米分布信息，需要优先使用与华北地区农业景观更相似的东北作物分布数据产品进行深度学习模型训练。在另一方面，我们发现相对于TOA（top of atmosphere）数据，采用SR数据更有利于模型在跨洲际尺度进行空间迁移。因此，在大尺度作物制图研究中，应优先考虑使用SR（surface reflectance）数据。本研究从一定程度上验证了影响农作物分类模型迁移性能的内在因素，为大尺度作物制图提供了一定的实验基础。

Timely and accurate global crop mapping is of great significance for global food security assessment. However, existing crop classification models are often targeted at specific regions, and their performance in other regions has not been fully evaluated. In order to realize the effective transfer of the model in large-scale regions, this study determined the critical period of crop growth in different regions, and filtered the remote sensing data during these critical periods so that the same crop in different regions showed similar characteristics on these remote sensing images. This helped the model achieve a better transfer effect. In this study, the MultiResUNet,　SegNet,　DeepLab V3+ and U-Net　models were trained using data from Northeast China, and the optimal F1 value for summer corn recognition in the study areas in North China can reach more than 0.97. In addition, this research analyzed the factors that affected the generalization ability of the model. The issues that this article focused on include: (1) Using existing crop distribution data products as the ground truth samples for model training to solve the problem of lack of training samples for the deep learning model. We compared and analyzed the applicability of the models trained using the US Cropland Data Layer (CDL) and Northeast crop distribution data products in North China. (2) We compared the generalization performance of depth models with different architectures. (3) We compared and analyzed the influence of different data type on the generalization ability of the model. (4) We have comparatively analyzed the impact of crop phenology changes on the generalization ability of the model. The results show that MultiResUNet has better generalization performance than other networks when the size of plot in training area and test area varies significantly However, the generalization ability of MultiResUNet alone still cannot completely overcome the adverse effect of the change of plot spatial morphology on model migration. In order to obtain more accurate information of maize distribution in North China, it is necessary to use the crop distribution data products in Northeast China that are more similar to the agricultural landscape in North China for deep learning model training. On the other hand, compared with TOA data, we found that SR data is more conducive to the spatial migration of the model at the trans-continental scale. Therefore, SR data should be given priority in large-scale crop mapping. This research provides a useful discussion for large-scale crop mapping research using only local samples.