西安电子科技大学学报 ›› 2021, Vol. 48 ›› Issue (2): 7-14.doi: 10.19665/j.issn1001-2400.2021.02.002

• 雷达技术进展专题 • 上一篇    下一篇

采用CNN-SSD的雷达HRRP小样本目标识别方法

郭泽坤1(),田隆1(),韩宁2(),王鹏辉1(),刘宏伟1(),陈渤1()   

  1. 1.西安电子科技大学 雷达信号处理国家重点实验室,陕西 西安 710071
    2.中国人民解放军32181部队,陕西 西安 710032
  • 收稿日期:2020-12-25 修回日期:2021-01-14 出版日期:2021-04-20 发布日期:2021-04-28
  • 通讯作者: 韩宁,陈渤
  • 作者简介:郭泽坤(1994—),男,西安电子科技大学博士研究生,E-mail: gzk1105@163.com|田 隆(1992—),男,西安电子科技大学博士研究生,E-mail: tianlong_xidian@163.com|王鹏辉(1983—),男,副教授,博士,E-mail: wangpenghui@mail.xidian.edu.cn|刘宏伟(1971—),男,教授,博士,E-mail: hwliu@xidian.edu.cn|郭泽坤(1994—),男,西安电子科技大学博士研究生,E-mail: gzk1105@163.com|田 隆(1992—),男,西安电子科技大学博士研究生,E-mail: tianlong_xidian@163.com|王鹏辉(1983—),男,副教授,博士,E-mail: wangpenghui@mail.xidian.edu.cn|刘宏伟(1971—),男,教授,博士,E-mail: hwliu@xidian.edu.cn
  • 基金资助:
    国家杰出青年科学基金(61525105);国家自然科学基金(61771361);高等学校学科创新引智计划(111计划)(B18039);国家杰出青年科学基金(61525105);国家自然科学基金(61771361);高等学校学科创新引智计划(111计划)(B18039)

Radar HRRP based few-shot target recognition with CNN-SSD

GUO Zekun1(),TIAN Long1(),HAN Ning2(),WANG Penghui1(),LIU Hongwei1(),CHEN Bo1()   

  1. 1. National Laboratory of Radar Signal Processing,Xidian University,Xi’an 710071,China
    2. Unit 32181 of PLA,Xi’an 710032,China
  • Received:2020-12-25 Revised:2021-01-14 Online:2021-04-20 Published:2021-04-28
  • Contact: Ning HAN,Bo CHEN

摘要:

雷达高分辨距离像非合作目标识别技术的发展主要受限于两个方面:一是由于非合作目标观测频率极低,导致带标签样本量严重不足,使非合作目标识别成为典型的小样本识别问题,这在学界依然是一个没有定论的开放性的热点和难点问题;二是现有的目标识别方法多基于完备数据集假设,使得其与非合作目标小样本目标识别问题严重失配。针对上述问题,对于非合作目标识别抛开数据集完备假设,提出了一种采用卷积神经网络模型连续自蒸馏的雷达高分辨距离像小样本目标识别方法。该方法首先利用包含45类合作目标的完备的训练数据集训练,得到一个初始的类别无关的特征提取器;基于此,进一步采用模型连续自蒸馏机制得到更具泛化能力的特征提取器;最后,在非合作目标上对所提取特征的泛化能力进行了测试。实验结果表明,对于5类非合作目标,所提方法在仅有1个、5个和10个训练样本的情况下,平均识别率分别达到61.26%,84.69%和92.52%,实现了对库外样本的快速、有效识别。

关键词: 雷达目标识别, 小样本学习, 特征提取, 高分辨距离像, 卷积神经网络, 连续自蒸馏, 雷达目标识别, 小样本学习, 特征提取, 高分辨距离像, 卷积神经网络, 连续自蒸馏

Abstract:

The development of radar high resolution range profile(HRRP)non-cooperative targets recognition technology is mainly limited by two aspects:(1) Due to the low observation frequency of non-cooperative targets,the number of labeled HRRPs is insufficient,making non-cooperative HRRP based target recognition a typical few-shot recognition problem,which is still a hot and difficult issue without definite conclusion in the academia.(2) The existing HRRP based target recognition methods are mostly based on the hypothesis of complete dataset,making them mismatch with non-cooperative target recognition in few-shot setting.In this paper,we put aside the complete hypothesis and propose an HRRP based few-shot target recognition method with CNN-SSD.The proposed method first uses a complete training HRRP containing 45 classes of cooperative targets to learn an initial category-independent feature extractor,on the basis of which we further utilize the model sequential self-distillation mechanism to obtain a more generalized feature extractor.Finally,the generalization ability of the extracted features is evaluated on unseen non-cooperative targets during training.Experimental results on self-simulated HRRP dataset reveal that the proposed method can achieve an average recognition rates of 61.26%,84.69% and 92.52% respectively when only 1,5 and 10 annotated HRRPs of non-cooperative targets are available.

Key words: radar target recognition, few shot learning, feature extraction, high resolution range profile, convolutional neural networks, sequential self-distillation, radar target recognition, few shot learning, feature extraction, high resolution range profile, convolutional neural networks, sequential self-distillation

中图分类号: 

  • TP183