›› 2019, Vol. 39 ›› Issue (1): 32-.doi: 10.16708/j.cnki.1000-758X.2018.0064

• 研究探讨 • 上一篇    下一篇

基于深度神经网络的空间目标结构识别

周驰,李智,徐灿   

  1. 1航天工程大学研究生院,北京101416 2航天工程大学航天指挥学院,北京101416
  • 出版日期:2019-02-25 发布日期:2019-02-25

Research on spatial target structure recognition based on deep neural network

ZHOU Chi,LI Zhi,XU Can   

  1. 1Graduate School,Space Engineering University,Beijing101416,China 2Academy of Space and Command,Space Engineering University,Beijing101416,China
  • Online:2019-02-25 Published:2019-02-25

摘要: 利用空间目标雷达散射截面(Radar Cross Section, RCS)序列开展空间目标结构识别是空间态势感知的重要组成部分。文章针对 RCS序列受目标物理特性、姿态特性影响大,序列信号非平稳特征明显的问题,利用深度神经网络(Deep Neural Network,DNN)算法解决空间目标结构特征识别的问题;针对特征提取不具区分度的问题,提出利用分形分析提取RCS序列的分数维特征,并利用Fisher判决率对传统特征进行选取;介绍了DNN算法以及数据处理过程;最后,利用一组仿真测试数据对算法进行了仿真验证。分析结果表明,DNN算法在解决利用RCS序列进行目标结构识别这一问题中具有鲁棒性强、识别准确的特点。

关键词: 深度神经网络, RCS序列, 结构识别, 分形分析, Fisher判决率

Abstract: The use of Radar Cross Section (RCS) sequences for spatial target structure recognition is an important part of space situational awareness. RCS sequence is easily affected by the target physical characteristics and attitude characteristics,and the nonstationary characteristics of the sequence signal are obvious. In this paper, deep neural network (DNN) algorithm was used to solve the problem of spatial target structural feature recognition. For the problem of feature extraction without distinguishing degree, fractal features were used to extract the fractal features of RCS sequences, and the Fisher′s decision rate was used for selecting traditional features. What′s more, the DNN algorithm and data processing process were introduced. Finally, a set of simulation test data were used to verify the algorithm. The analysis results show that the DNN algorithm is robust and accurate in solving the problem of using RCS sequence to identify the target structure.

Key words: Deep Neural Network, Radar Cross Section sequence, structural identification, fractal analysis, Fisher decision rate