Research And Application Of EM Optimization Algorithm Based On GMM

With the continuous development of the field of artificial intelligence,the application of cluster analysis is more and more extensive.Its main function is to classify the data reasonably,and then classify the data into different clusters.The Gaussian mixture model(GMM)learns some probability density functions,and the probability that each sample data point is divided into clusters is represented by the probability.This clustering method is called soft clustering..GMM is mainly used in intelligent traffic control systems,moving target detection,image recognition,assisted driving systems and other fields.The GMM estimates the probability density distribution of the sample and uses the Expectation Maximization(EM)training.However,the EM algorithm of the Gaussian mixture model still has the deficiency of the algorithm itself,which seriously affects the quality of the clustering.In this paper,there are two main disadvantages of the EM algorithm for Gaussian mixture model,namely the data initialization sensitivity problem and the local convergence problem,and the corresponding improvement methods are proposed.First of all,for the Gaussian mixture model sensitive to the initial value of the data,this paper adopts the hierarchical clustering algorithm to preprocess the Gaussian mixture model parameters.Secondly,when estimating the mixed model parameters by the EM algorithm of Gaussian mixture model,the estimated parameters obtained are highly localized optimal solutions,affecting the final clustering results of the Gaussian mixture model.This paper uses the approximate skeleton theory to make up for the lack of local convergence.The approximate skeleton can capture multiple local optimal solutions.By applying the approximate skeleton to the design of the clustering algorithm,the global optimal solution can be solved,thereby avoiding the problem of local convergence when the GMM processes a large batch of data sets.The distribution of the sample dataset will be best fitted.Finally,this paper puts the clustering algorithm optimized by EM algorithm based on Gaussian mixture model into the concrete application of spatial index.The application of spatial indexing technology is becoming more and more widespread.R*-tree,as an important spatial index structure,still has the defect that the minimum overlapping rectangles overlap.This paper uses the optimized Gaussian mixture model clustering algorithm to reconstruct R*-tree,effectively reduces the overlap ratio of the minimum outsourced rectangle,strengthens the similarity between attributes,and reduces the number of paths to the target object.At the same time,the search time is shortened and the spatial indexing efficiency is improved.