一种新型自适应估计高斯混合模型阶数的快速EM算法  

A Novel Estimation of Gaussian Mixture Models on EM Algorithm

作　　者：戴卿[1,2] 

机构地区：[1]重庆水利电力职业技术学院建筑工程系,重庆402100 [2]信息工程大学空间地理信息学院,河南郑州450000

出　　处：《兰州工业学院学报》2017年第1期59-63,共5页Journal of Lanzhou Institute of Technology

基　　金：渝水职院重点项目(K201514;K201510);重庆教委科学研究项目(KJ1735452)

摘　　要：高斯混合模型作为一种良好的非高斯噪声处理技术,应用领域广泛,其参数估计常采用通过基于最大似然估计的期望最大化迭代算法(Expectation Maximization,EM).针对常规EM算法性能受迭代初值和模型阶数的影响,且计算量较大的问题,提出了一种在线自适应估计模型阶数的快速EM算法,该算法由单高斯模型逐步分离出多高斯分量.数据仿真实验表明,新算法初值设置简单,避免了局部收敛,有效提高了计算效率.Gaussian mixture model is considered as a nice non-Gaussian distribution data processing technology which are widely used in many different applications. And the parameter estimation is based on the Maximum Likelihood Estimation（EM） algorithm. However,the performance of the conventional EM algorithm is affected by initial value and number of mixture components,the computational complexity is larger. This paper discusses a fast EM algorithm for on-line estimating the number of mixture components. This algorithm separates the Gaussian components step by step from the single Gaussian model. The simulation results show that the new algorithm has merits of simpleness,avoiding local convergence as well as computational efficiency.

关 键 词：最大似然估计 高斯混合模型 EM算法 模型阶数 时间复杂度 

分 类 号：TN96[电子电信—信号与信息处理]