| Gaussian mixture model(GMM)is widely used in data clustering,image segmentation,and other fields.Because the hidden variables in the composition are introduced and the number of components in the Gaussian mixture model also needs to be estimated in advance,the problem of parameter estimation cannot be solved using maximum likelihood estimation.The expectation maximization(EM)algorithm is a common tool for estimating the parameters of Gaussian mixture models(GMM).However,it is highly sensitive to initial value and easily gets trapped in a local optimum.In this paper,a new iterative initialization method-MRIPEM random initialized strategy was proposed,which incorporates the ideas of multiple restarts,iterations and clustering.First of all,the mean vector and covariance matrix of sample are calculated as the initial values of the iteration.Then,the optimal feature vector is selected from the candidate feature vectors by the maximum Mahalanobis distance as a new partition vector for clustering.The parameter values are renewed continuously according to the clustering results.Due to the fact that the initial iteration value of the algorithm and the parameter values determined during the algorithm process are completely calculated from the sample observations,the randomness is reduced to a certain extent and the stability of the results is increased.On the other hand,this paper uses Mahalanobis distance to select the candidate vector farthest from all determined center vector points as the partition vector,which reduces the probability of samples being misclassified.In addition,the algorithm proposed in this article can also customize parameters according to user needs,and select different initialization levels.Therefore,MRIPEM algorithm reduces the possibility of EM algorithm falling into a local optimal value in principle.In order to verify the effectiveness and applicability of the MRIPEM algorithm,the experiments on simulated and real datasets were conducted in this paper,and the method of MRIPEM was compared with two other popular random initialization strategies: em EM and Rnd EM.The experimental results show that the MRIPEM initialization method has strong application for well separated low dimensional datasets. |
