高分辨率雷达以低擦地角观测粗糙海表面时杂波幅度明显增强,产生海尖峰效应。海尖峰与甲稳海杂波的统计特性差别显著,使用单一概率密度函数(PDF)的传统统计建模方法难以精确描述回波特性,尤其在回波中包含目标信号时,这种不适应更为严重。该文将连续型隐马尔可夫模型(CHMM)用于海杂波建模,把海面回波分为甲稳海杂波、海尖峰和目标回波3个状态,使用高斯混合密度模型(GMDM)建立各状态观测值的连续PDF表达式,使用Baum-Welch算法对CHMM的参数进行计算和重估。同时,修正了基于GMDM的CHMM观测值状态联合概率公式,解决了GMDM参数迭代求解过程中的分母下溢出问题,为海杂波建模与分析提供了一种新的方法。最后对实际雷达采集数据的分析证明了该方法的有效性。
When high resolution radar observes rough sea surface with low grazing angle the level of sea clutter increases obviously and the sea spike phenomena appears. It is not accurate to analyze the radar echo including stationary sea clutter, spike and target with the conventional statistical model based on single Probability Density Function (PDF). In this paper, the Continuous Hidden Markov Model (CHMM) is used to model and analyze the sea clutter. The echoes from sea surface are divided into three states of CHMM stationary sea clutter, sea spike and target. The Gaussian Mixture Density Model (GMDM) and the Baum-Welch algorithm are used to construct the PDF expressions of the observations of the three states and re-estimate the CHMM parameters, respectively. At the same time, the expressing of the observation-state joint probability is modified to avoid the underflow of the denominator during the iterative procedure of the GMDM parameters. Thus, a new method is proposed for modeling and analyzing sea clutter.