基于统计建模的字典学习算法在HRRP的应用

基于字典学习模型能真实反映雷达高分辨距离像(radar high resolution range profile, HRRP)潜在结构特征和统计建模算法可有效解决HRRP姿态敏感性问题的特点,运用统计建模划分HRRP角域,对鉴别字典的原子选取和判别优化问题开展研究。首先提出了基于概率主分量分析的最大概率差值算法,自适应划分HRRP角域获取帧界线。其次,利用帧界线对应功率谱特征构成初始化鉴别字典,在鉴别字典基础上优化判别准则,引入原子稀疏相似误差约束最优字典更新实现测试样本分类。雷达实测数据的实验结果验证了该算法可提高目标识别率,同时对噪声干扰具有很好的鲁棒性。

Application of dictionary learning algorithm in HRRP based on statistical modeling

The dictionary learning model can really reflect radar high resolution range profile (HRRP) potential structural characteristics and the statistical modeling algorithm can effectively solve the HRRP attitude sensitivity problem. Based on those features, researches on the selection of atoms and the discriminant optimization problem for label consist K-singular value decomposition （LC-KSVD） have been carried out by using statistical modeling in dividing the HRRP’s angular domain. Firstly, the maximum probability difference algorithm based on probabilistic principal component analysis is proposed to adapt HRRP angular domain to obtain the frame boundary. Secondly, based on LC-KSVD, the discriminant criterion is constructed by using the power spectrum of the frame boundary and the introduction of atomic sparse similarity error constraint in optimal dictionary selection to clarify test samples. The experimental results of the radar data show that this algorithm can improve the target recognition rate, and has good robustness to the noise interference.