基于混合概率主成分分析的HRRP特征提取

针对主成分分析(principal component analysis, PCA)等数据压缩方法用于雷达高分辨距离像(high resolution range profile, HRRP)的特征提取,只能反映固定方位帧内HRRP线性结构,而无法准确描述目标,导致识别性能下降的问题,提出了一种基于混合概率PCA的HRRP特征提取方法。该方法利用期望最大值(Expectation maximization, EM)算法求解HRRP的一、二阶统计参数,能够真实反映数据分布,以分布趋同的原则实现不同方位帧的聚类,减少模板数量。最后通过自适应高斯分类器和Kullback-Leibler距离分类器识别获取的统计特征,可进一步改善识别性能。仿真实验验证了该方法能够在降低数据维数的同时,实现HRRP统计特征的提取,能一定程度上削弱方位敏感性的影响。

HRRP feature extraction based on mixtures of probabilistic principal component analysis

Using the data compression technology like principal component analysis (PCA) for high resolution range profile (HRRP) feature extraction will lead to the decrease of recognition rate, for it only reflects the linear structure of HRRP in fixed azimuth frames which cannot accurately describe the target. So the mixtures of probabilistic PCA method is proposed for solving this problem, and the expectation maximization algorithm is adopted to estimate the statistic parameters for this method. Because the real data distribution can be obtained from the algorithm, this method would offer the potential to model the similar density of HRRP adequately for clustering to separate azimuth frame and reduce the local dimensionality for storing the template. Finally, the adaptive Gaussian classifier (AGC) and Kullback-Leibler (KL) distance classifier are both utilized to test the performance of obtained statistical features. Simulation experimental results show that this method not only reduce the dimensionality of HRRP, but also extract the statistical feature to eliminate the azimuth sensitivity.