Band Selection with Bhattacharyya Distance Based on the Gaussian Mixture Model for Hyperspectral Image Classification

Abstract

This paper investigate a new band selection approach with the Bhattacharyya distance based on the Gaussian Mixture model (GMM) for Hyperspectral image classification. Our main motivation to model the Bhattacharyya distance using GMM is due to the fact that this tool is well known for capturing non-Gaussian statistic of multivariate data and that is less sensitive to estimation error problem than purely non parametric models. To estimate the parameters of GMM, a Robust Expectation-Maximization (REM) algorithm is used. REM solves the shortcoming of the classical Expectation-Maximization (EM) algorithm by dynamically adapting the number of clusters to the data structure. The selected bands with the proposed approach are compared, in terms of classification accuracy, to the Bhattacharyya expressed in its parametric form and the Bhattacharyya modelled with GMM using the classical EM algorithm. The experiment were carried out on two real hyperspectral images, the Indiana Pines (92AV3C) sub-scene and the Kennedy Space Center (KSC) dataset, and the experimental results have demonstrated the effectiveness of our proposed method in terms of classification accuracy with fewer bands.