准确的湿地分类可掌握湿地时空分异特征，在湿地研究中占据重要地位。针对现有基于小样本学习的湿地分类方法仅局限于利用目标域或单源域数据的问题，本文提出一种3D多源域自注意力小样本学习模型3D-MDAFSL（3D Multi-source Domain self-Attention Few-Shot Learning）。首先，结合卷积和注意力机制的优势，设计基于自注意力机制和深度残差卷积的3D特征提取器；然后，采用对抗域自适应策略实现多源域特征对齐，在每个域分别进行小样本学习；最后，利用训练好的模型提取特征，并将特征输入至K近邻（K-nearest Neighbor）分类器以获取分类结果。结果表明，3D特征提取器相比无特征提取框架的湿地总体分类精度提升约6.79%；当使用多源域数据集时，3D-MDAFSL模型对中山市Sentinel-2湿地数据集的总体分类精度能达到93.52%，相比于现有算法有明显提升。本文所提出的3D-MDAFSL模型在湿地地物高精度提取和分类中有较好的应用价值。

Accurate wetland classification methods can quickly grasp the spatial-temporal variation characteristics of wetlands, and plays an important role in wetland research. Aiming at the problem that the existing wetland classification method based on few-shot learning is limited to the use of target domain or single source domain dataset, this paper proposes a 3D multi-source domain self-attention few-shot learning model 3D-MDAFSL (3D Multi-source Domain self-Attention Few-Shot Learning). Firstly, combining the advantages of convolution and attention mechanism, a 3D feature extractor based on self-attention mechanism and deep residual convolution is designed. Then, the conditional adversarial domain adaptation strategy is used to achieve multi-source domain distribution alignment, and few-shot learning is performed separately in each domain. Finally, the features extracted by the trained model are input to the K-nearest Neighbor classifier to obtain classification results. The results show that compared with the framework without feature extraction, the 3D feature extractor improves the overall accuracy by about 6.79%. When using multi-source domain datasets, the overall accuracy of the 3D-MDAFSL model for the Sentinel-2 wetland dataset in Zhongshan City can reach 93.52%, which is significantly improved compared with the existing algorithms. The 3D-MDAFSL model proposed in this paper has good application value in the high-precision extraction and classification of wetland features.