电子学报 ›› 2020, Vol. 48 ›› Issue (6): 1124-1131.DOI: 10.3969/j.issn.0372-2112.2020.06.012

• 学术论文 • 上一篇    下一篇

小样本条件下基于数据增强和WACGAN的雷达目标识别算法

朱克凡, 王杰贵, 刘有军   

  1. 国防科技大学电子对抗学院, 安徽合肥 230037
  • 收稿日期:2019-06-18 修回日期:2019-11-07 出版日期:2020-06-25 发布日期:2020-06-25
  • 作者简介:朱克凡 男,1994年出生,山东青岛人.现为国防科技大学硕士研究生,主要研究方向为雷达目标识别.
    王杰贵 男,1969年生,现为国防科技大学电子对抗学院副教授、硕士生导师,研究方向为雷达信号处理.
    刘有军 男,1975年生,现为国防科技大学电子对抗学院副教授,研究方向为电子对抗信息处理和装备计量技术研究.
  • 基金资助:
    国防预研基金(No.9140C100404120C1003)

Radar Target Recognition Algorithm Based on Data Augmentation and WACGAN with a Limited Training Data

ZHUKe-fan, WANG Jie-gui, LIU You-jun   

  1. Electronic Countermeasure Institute of National University of Defense Technology, Hefei, Anhui 230037, China
  • Received:2019-06-18 Revised:2019-11-07 Online:2020-06-25 Published:2020-06-25

摘要: 目前小样本条件下高分辨距离像雷达目标识别算法存在识别率较低、识别率稳定度较差等问题,对此,本文提出了基于数据增强和加权辅助分类生成对抗网络(Weighted Auxiliary Classifier Generative Adversarial Networks,WACGAN)的雷达目标识别算法.该算法首先根据雷达目标散射特性,通过时间镜像数据增强方法扩充数据集,然后将扩充数据集输入WACGAN,通过自动选择高质量的生成样本,使判别器在标签样本监督学习的基础上得到进一步优化,最后直接利用判别器实现对雷达目标的有效识别.仿真实验结果表明,本文算法在不增加识别时间的基础上,有效提高了小样本条件下对雷达目标的识别率和识别稳定度.

关键词: 雷达目标识别, 数据增强, 生成对抗网络

Abstract: At present,the existing high-resolution range profile radar target recognition algorithm with limited training data still has several drawbacks (e.g.,low recognition accuracy and poor recognition stability).To this end,an efficient radar target recognition algorithm is developed in this paper,which is based on data augmentation and Weighted Auxiliary Classifier Generative Adversarial Networks (WACGAN).Specifically,we expand data set by using the data augmentation method based on time mirroring,and the radar target scattering characteristics is considered.After that,the WACGAN with expanded data set is used to automatically select high-quality generated samples and further optimize the discriminator,which has been improved through the supervised learning.Then,the optimized discriminator is used to recognize radar target.Finally,several numerical experiments have been carried out to demonstrate that,under the condition of limited training data,the proposed algorithm possesses higher recognition accuracy and better recognition stability without increasing recognition time.

Key words: radar target recognition, data augmentation, generative adversarial network

中图分类号: