宇航学报

宇航学报 ›› 2012, Vol. 33 ›› Issue (9): 1269-1278.doi: 10.3873/j.issn.1000-1328.2012.09.012

• 电子信息 • 上一篇    下一篇

基于时频矩阵非负分解特征的多视角SAR目标识别

张新征1, 刘书君1, 黄培康2   

  1. （1. 重庆大学通信工程学院，重庆 400044； 2. 中国航天科工集团科技委，北京 100854）
  • 收稿日期:2011-09-05 修回日期:2012-05-31 出版日期:2012-09-15 发布日期:2012-09-13
  • 基金资助:
    中央高校基本科研业务费资助（CDJRC11160003）；中国博士后基金资助（2011M501389）

Multi\|Perspective SAR Target Recognition Based on Non\|Negative Time\|Frequency Matrix Factorization

ZHANG Xinzheng1, LIU Shujun1, HUANG Peikang2   

  1. （1. College of Communication Engineering, ChongQing University, Chongqing 400044, China;

    2. The Science Committee of China Aerospace Science & Industry Corporation, Beijing 100854, China)

  • Received:2011-09-05 Revised:2012-05-31 Online:2012-09-15 Published:2012-09-13

PDF

923

摘要： 针对多视角合成孔径雷达(Synthetic Aperture Radar, SAR)目标识别问题，提出一种基于目标高分辨率距离像(High Range Resolution Profile，HRRP)时频矩阵非负分解特征提取和识别方法。该方法首先对SAR图像进行滤波预处理，得到相应的目标HRRP序列；然后采用匹配追踪时频分析方法计算得到目标HRRP的时频矩阵；应用非负矩阵分解技术分解时频矩阵，得到相应的谱矢量和时相矢量。基于分解得到的谱矢量和时相矢量提取时频域矩特征和稀疏特征。最后，应用隐马尔科夫模型(Hidden Markov Model, HMM)对这些时频特征序列建模及识别。采用美国运动和静止目标获取与识别(Moving and Stationary Target Acquisition and Recognition, MSTAR)计划公开发布的SAR目标数据库的实验结果表明，该方法不仅能有效降低时频域特征的维数，而且识别性能优于传统的时频域特征。

关键词: SAR, 时频分析, 非负矩阵分解, 特征提取, 目标识别

Abstract: A new approach is developed for multi\|perspective synthetic aperture radar (SAR) target feature extraction and classification based on non\|negative time\|frequency matrix decomposition of high range resolution profiles (HRRPs). The developed approach consists of following stages: SAR images are converted into HRRPs through several filtering operations and the time\|frequency matrix of HRRP is constructed by using the matching\|pursuit time\|frequency analysis technique firstly. Then the time\|frequency matrix of HRRP is decomposed into its spectral and temporal component vectors by applying the non\|negative matrix factorization technique. The time\|frequency moment features and sparsity features are extracted from spectral and temporal component vectors. Finally, hidden markov models (HMM) are utilized to model and classify time\|frequency feature sequences. Experimental results with the Moving and Stationary Target Acquisition and Recognition (MSTAR) data show that the proposed approach not only decreases feature dimensions significantly, but also improves the classification accuracy compared to traditional time\|frequency features.

Key words: SAR, Time-frequency analysis, Non-neGative matrix factorization, Feature extraction, Target recoGnition

中图分类号: