【摘要】 遥感影像分类与识别是近年来深度学习以及图像分类与识别研究的热点，其中一个关键问题是因样本数据集的数据较少而极易出现过拟合。许多图像分类的模型和方法并不完全适用于遥感影像分类，将小样本学习与遥感影像处理结合起来，实现遥感影像数据增强和识别模型优化是一个可行的思路。根据小样本学习的发展现状，针对特征提取、模型分类方法，归纳总结了典型学习方法的原理及其在相关领域的应用；分析遥感影像处理的现状和存在问题，基于适用场景、优缺点对各方法进行了比较；通过分析小样本学习在高分遥感影像分类与识别上的应用，发现引入注意力机制和迁移学习后，小样本学习能够用于样本数据量小的遥感影像分类。更多还原

【Abstract】 Remote sensing image classification and recognition has been a hot topic in deep learning and image classification and recognition research in recent years. One of the key issues is that the sample data scale is small and prone to over-fitting. Therefore, many image classification models and methods cannot be fully applied to remote sensing image classification. It is a practicable idea to combine few-shot learning with remote sensing image processing to solve the problem of remote sensing image samples by data augmentation and optimizing the recognition model. According to the development of few-shot learning, the principles of typical learning methods and their applications in related fields are summarized for feature extraction and model classification methods. The status and problems of remote sensing image processing are analyzed, and the methods are compared based on the applicable scenarios, advantages and disadvantages. Through the analysis of the application of few-shot learning in high score remote sensing image classification and recognition, it is found that after the attention mechanism and transfer learning are introduced, few-shot learning can be used for remote sensing image classification with small sample data.更多还原