混合高斯参数估计的动态簇算法

Dynamic cluster algorithm for Gaussian mixture parameter estimation

混合高斯概率密度模型可以很好地拟合非高斯样本的概率密度。在各高斯分量概率密度互不重叠的条件下,使用动态簇算法可以快速而精确地估计出混合高斯概率密度模型参数。这是一种基于最小均方差原则的递推算法,在正向推导出各种可能的簇边界后,再根据确定的最末边界值逆向推定各前导簇边界,从而得到混合高斯概率密度模型参数估计值。描述模型及参数估计问题之后,动态簇算法被推导出来。然后深入探讨了该算法的实质及适用条件。最后结合数值仿真实例,分析了动态簇算法的估计性能。

Probability density of non-Gaussian processes can be well fit by Gaussian mixture model whose parameters can be estimated through the dynamic clutter algorithm that is a recursion on the principle of minimum mean-square deviation. All possible boundaries of each clutter are forward-derived. Since the right boundary of final clutter is determinate, boundaries of previous clutters can be recurred backwards one by one. Thus the Gaussian mixture parameters are estimated. After descriptions of the model and the estimation problem, dynamic clutter algorithm for Gaussian mixture parameters is obtained. Its essential ideas and applicability are discussed in detail. A numerical example is presented to study the performance of estimation.