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.
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)
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
maximum likelihood estimation
Gaussian mixture mode
EM algorithm
model order
time complexity
