西安电子科技大学学报 ›› 2024, Vol. 51 ›› Issue (5): 46-57.doi: 10.19665/j.issn1001-2400.20240602

• 信息与通信工程 • 上一篇    下一篇

基于自组织神经网络的EVD杂波抑制算法

史家琪1(), 杨明磊1(), 连昊1(), 叶舟2(), 徐光辉2()   

  1. 1.西安电子科技大学 雷达信号处理全国重点实验室,陕西 西安 710071
    2.上海航天电子通讯设备研究所,上海 201109
  • 收稿日期:2023-12-20 出版日期:2024-07-04 发布日期:2024-07-04
  • 通讯作者: 杨明磊(1981—),男,教授,E-mail:mlyang@mail.xidian.edu.cn
  • 作者简介:史家琪(1999—),男,西安电子科技大学硕士研究生,E-mail:jiaqs0102@163.com
    连 昊(1996—),男,西安电子科技大学博士研究生,E-mail:haolian1224@163.com
    叶 舟(1988—),男,高级工程师,E-mail:yz101011@126.com
    徐光辉(1987—),男,高级工程师,E-mail:xghuuuuuui@163.com
  • 基金资助:
    国家自然科学基金(62171336);高等学校学科创新引智计划(111计划)(STAST2020086)

EVD clutter suppression method based on the self-organizing neural network

SHI Jiaqi1(), YANG Minglei1(), LIAN Hao1(), YE Zhou2(), XU Guanghui2()   

  1. 1. National Key Laboratory of Radar Signal Processing,Xidian University,Xi’an 710071,China
    2. Shanghai Aerospace Electronic Communication Equipment Research Institute,Shanghai 201109,China
  • Received:2023-12-20 Online:2024-07-04 Published:2024-07-04

摘要:

强杂波环境下慢速运动目标的杂波抑制一直是雷达领域的研究难点,通过子空间分解法来抑制杂波是一种常用的方法,但传统子空间分解法依赖于过往经验选取杂波基、自适应性差。基于K-均值聚类的SVD杂波抑制算法弥补了上述缺陷,然而当慢速运动目标与杂波在多普勒谱上接近或混叠时,这种算法的特征集区分度大幅下降,聚类结果变得不稳定。为此提出了一种基于自组织神经网络的特征值分解杂波抑制算法。首先,深入分析慢速运动目标和杂波、噪声的差异,利用回波信号矩阵特征值分解后得到的特征值和特征向量,提取针对慢速运动目标和杂波区分度高的特征来构建特征集。其次,采用受初始值影响小、聚类结果稳定的自组织神经网络进行聚类,自适应选取构造杂波子空间的杂波基,最后通过正交子空间投影来抑制杂波。仿真和实测数据结果表明该算法能有效抑制强静止杂波和慢速杂波,实现对慢速运动目标的检测,算法具有较强的稳健性和工程实用性。

关键词: 慢速运动目标, 杂波, 特征值分解, 自组织神经网络

Abstract:

The subspace decomposition method is a common method for clutter suppression of slow moving targets in strong clutter environment.But the traditional subspace decomposition method has a poor adaptability.The SVD clutter suppression algorithm based on K-means clustering makes up for the above defects,but when the slow-moving target is close to the clutter Doppler or aliasing,the feature set discrimination decreases and the clustering results are unstable.Therefore,an eigenvalue-decomposition(EVD) clutter suppression algorithm based on self-organizing neural networks is proposed,with the differences between targets,clutter and noise analyzed deeply,and the features with high differentiation between slow-moving targets and clutter extracted to construct the feature set.Then,the self-organizing neural network,which is less affected by the initial value and has stable clustering results,is used for clustering,adaptive selection of clutter basis to construct clutter subspace.Finally,the clutter is suppressed by orthotropic subspace projection.Simulation and measured data are used to verify the performance of the algorithm.By combining with the target tracking algorithm,it is further verified that the algorithm has strong robustness and engineering practicability.

Key words: slow-moving target, clutter, eigen value decomposition(EVD), self-organizing neural network

中图分类号: 

  • TN957.51