系统工程与电子技术 ›› 2022, Vol. 44 ›› Issue (12): 3595-3602.doi: 10.12305/j.issn.1001-506X.2022.12.02

• 电子技术 • 上一篇    

基于集成深度学习的有源干扰智能分类

吕勤哲1, 全英汇1,*, 沙明辉2, 董淑仙1, 邢孟道3   

  1. 1. 西安电子科技大学电子工程学院, 陕西 西安 710071
    2. 北京无线电测量研究所, 北京 100854
    3. 西安电子科技大学前沿交叉研究院, 陕西 西安 710071
  • 收稿日期:2021-05-10 出版日期:2022-11-14 发布日期:2022-11-24
  • 通讯作者: 全英汇
  • 作者简介:吕勤哲 (1998—), 男, 博士研究生, 主要研究方向为雷达干扰识别对抗|全英汇 (1981—), 男, 教授, 博士, 主要研究方向为雷达实时信号处理|沙明辉 (1986—), 男, 研究员, 博士, 主要研究方向为雷达抗干扰和信号处理|董淑仙 (1995—), 女, 博士研究生, 主要研究方向为雷达信号处理|邢孟道 (1975—), 男, 教授, 博士, 主要研究方向为SAR/ISAR成像、动目标检测
  • 基金资助:
    国家自然科学基金(61772397);国家重点研发计划(2016YFE0200400);陕西省科技创新团队(2019TD-002)

Ensemble deep learning-based intelligent classification of active jamming

Qinzhe LYU1, Yinghui QUAN1,*, Minghui SHA2, Shuxian DONG1, Mengdao XING3   

  1. 1. School of Electronic Engineering, Xidian University, Xi'an 710071, China
    2. Beijing Institute of Radio Measurement, Beijing 100854, China
    3. Academy of Advanced Interdisciplinary Research, Xidian University, Xi'an 710071, China
  • Received:2021-05-10 Online:2022-11-14 Published:2022-11-24
  • Contact: Yinghui QUAN

摘要:

针对现有基于机器学习的雷达有源干扰分类大多需要构建人工特征集且小样本情况下分类精度低的问题, 提出一种基于多通道特征融合的集成卷积神经网络(convolutional neural network, CNN)分类方法。首先, 建立多种有源干扰的数学模型, 仿真并利用短时傅里叶变换获得其时频分布图; 其次, 提取时频分布图的实部、虚部和模值三通道特征, 通过多种特征组合方式建立不同特征组合的样本集; 最终, 构建以CNN为基分类器的集成深度学习模型, 每个CNN分别提取不同样本集的特征, 对所有基分类器的预测结果做多数投票得到集成模型的整体预测结果。实验表明, 该方法能够有效实现小样本情况下多类有源干扰的高精度智能化识别。

关键词: 有源干扰分类, 短时傅里叶变换, 集成学习, 卷积神经网络, 小样本

Abstract:

Aiming at the problems that most existing machine learning-based classification of radar active jamming need to construct artificial feature sets and low classification accuracy in the case of small samples, an ensemble convolutional neural network (CNN) classification method based on multichannel feature fusion is proposed. Firstly, multiple mathematical models of active jamming are established, simulated and the corresponding time-frequency profiles are obtained by using the short-time Fourier transform (STFT). Secondly, real part, imaginary part, and modulus three-channel features of time-frequency distribution plots are extracted to establish sample sets containing different combinations of features through multiple feature combinations. Ultimately, an ensemble depth model with CNN as the base classifier is constructed, each CNN separately extracts the features of different sample sets, and majority voting on the prediction results of all base classifiers gives the overall prediction results of the ensemble model. The experiments show that the proposed method can effectively realize highly accurate intelligent identification of multiple classes of active jamming in the case of small samples.

Key words: active jamming classification, short-time Fourier transform (STFT), ensemble learning, convolutional neural network (CNN), small sample

中图分类号: