小样本图像识别研究综述

Research on the Application of Few-Shot Learning in Image Recognition

摘要:

随着大规模数据集的发展,基于深度学习的神经网络模型在人脸识别、智能驾驶、工业质量检测、医疗诊断等图像识别领域取得了优异的表现。然而在实际场景的推广应用中,由于诸多因素的限制,研究人员无法获取大量的满足要求的样本数据,难以达到满意的识别效果,因此进行小样本情形下的图像识别研究是十分有意义的,系统地梳理了近年来小样本学习在图像识别领域的应用进展,主要从基于数据增强、基于表征学习以及基于学习策略等三个方面对相关工作进行了介绍分析,并根据目前的研究情况,展望探讨了未来的研究方向。

Abstract:

With the development of large-scale datasets, deep learning-based neural network models have achieved excellent results in image recognition fields such as face recognition, intelligent driving, and medical diagnosis. However, in practical applications, due to various limiting factors, researchers are unable to obtain a large number of samples that meet the requirements. Therefore, studying image recognition under small sample conditions is very meaningful. This work systematically reviews recent research progress in few-shot learning in the field of image recognition, introducing and analyzing related work from three aspects: data augmentation-based methods, representation learning-based methods, and learning strategy-based methods. Based on the current research situation, it also explores and discusses future research directions.