Hyper spectral image classify by using spectral –spatial classification based on Extended random walker (ERW)s .In this ERW have mainly two steps . First go to pixel wise classification by using support vector machine (SVM) .It is used for classification probability maps for a hyper spectral images. The probabilities of hyper spectral Pixel belongs to different classes .The second approach is obtain pixel wise probability maps are optimized by Extended random walker algorithm. Based on three factors i.e, Pixel wise statistics information by SVM classifier, spatial correlation among adjacent pixels modeled by the weights of graph edges and the connectedness between the training and test samples modeled by random walkers used to the class of the test pixel determined .So these three factors considered in ERW. By using Gaussian mixture model method in the proposed method shows very good classification and high accuracy performs for three widely used real hyper spectral data sets even the number of training samples is relatively small