面向智能感知的小样本学习研究综述

小样本学习指只利用目标类别的少量监督信息来训练机器学习模型。由于其实用价值，学术界和工业界提出很多针对该问题的解决方案，但是目前国内缺少该问题的综述。本文对国内外学者提出的小样本学习算法及基于小样本学习的目标检测算法进行了系统的总结和探索。首先，给出了小样本学习的问题定义，列举其与其他一些经典的机器学习问题之间的联系，同时从理论上阐述小样本学习问题面临的挑战；接着，对基于小样本学习的图像分类进行了概述，并对其中代表性的工作进行介绍与分析；在此基础上，重点针对基于小样本学习的目标检测，特别是零样本条件下的目标检测问题，详细介绍和分析了现有的研究工作；最后，立足于现有方法的优缺点，从问题设定、理论研究、实现技术以及应用场景等几个方面对小样本学习的未来发展进行了展望，期望为该领域后续的研究工作提供启示。

A survey of few shot learning based on intelligent perception

Few-shot learning refers to using only a small amount of supervision information of the target class to train the machine learning model. Due to its practical values, recent advances in few-shot learning by academia and industry have made significant contributions. However, there were few reviews on this issue in China. This paper systematically summarizes and explores the few-shot learning algorithms and the object detection algorithms based on few-shot learning. Firstly, the problem definition of few-shot learning is given, and its connections with other classic machine learning problems are also enumerated. Meanwhile, the theoretical challenges of the problem of few-shot learning are explained. Then, we summarize the image classification based on few-shot learning, and analyze its representative works. Based on this, we focus on the problem of few-shot object detection, especially the problem of zero-shot object detection, and analyze the existing research works in detail. Finally, we look forward to the future development of few-shot learning in terms of problem setting, theoretical research, implementation technology, and application scenarios based on the advantages and disadvantages of the existing methods. It is expected to provide inspirations for the subsequent research works in this field.