An Instructed Unsupervised Classification Method for Remote Sensing Image Based on Gaussian Mixture Model

We propose a new unsupervised classification method,GMM-UC,for remote sensing image based on Gaussian mixture model.This method is based on the theory of finite mixture model that the remote sensing data is mixed by a finite number of sub Gaussian distributions to a certain percentage.Through the improved EM algorithm,GMM-UC automatically determines the number of sub-Gaussians and its parameters,then restores all land objects (a sub-Gaussian is corresponding with a class of object).The improved EM algorithm solves the problems of high dependence on the initialization parameters and easily converging to the boundary of parameter space in standard EM algorithm effectively.Due to the adaptive design of the number of sub-Gaussian,the EM algorithm can find the shapes of the data distributions in a feature space furthest at a certain range,and the sub-Gaussians and and their mixed distributions is considered optimal.Experiments indicate that this method overcomes the fault of traditional unsupervised classification method to some extent,and expands the application field of it.