基于迁移学习的小样本目标识别研究进展与展望

Research Progress and Prospect of Small Sample Target Recognition Based on Transfer Learning

在空、天、海等复杂环境下的目标识别任务中,高质量的样本数据往往较少。特别是在干扰对抗环境下,某些特定领域的目标信息获取困难,可靠的标注数据较少。小样本问题对深度学习技术在目标识别任务中的应用提出了新的挑战。迁移学习为小样本不确定环境下的目标识别问题提供了新的研究思路。本文针对小样本目标问题,以机载雷达等空天传感器信息对海面目标识别为例,介绍了迁移学习的主要思路和方法,对迁移学习在海面目标识别问题中的应用现状进展进行了总结;分析和归纳了迁移学习在海面目标识别应用中的主要挑战。最后对可解释性及鲁棒性的海洋目标识别技术需求及未来发展方向进行了展望。

In the task of target recognition under complex environment such as sky,space and sea,there are often less high-quality sample data.Especially in the context of interference and countermove,it is difficult to obtain target information in some specific fields,and there are few reliable labeled data.The small sample problem brings new challenges to the application of deep learning technology in target recognition.Transfer learning provides a new research idea for target recognition under small sample and uncertain environment.Aiming at the problem of small sample target,this paper introduces the main ideas and methods of transfer learning and summarizes the progress of the application of transfer learning in sea target recognition by taking airborne radar and other airborne sensor information as an example.The main challenges of transfer learning in sea target recognition are analyzed and summarized.Finally,the requirement of technology and future direction of development for interpretable and robust sea target recognition are prospected.