A Unique Spectral Spatial Bayesian Framework via Elastic Net Regression for the Classification of Hyperspectral images

Abstract

This article presents not to simply a unique two stage regularization/shrinkage estimator for regression; rather, explicit to make the Bayesian framework connection to the Elastic Net procedure via the post-processed Edge preserving filtering which consist of two steps. We evaluated the quality of bands with pixel-based classifier associated with the Elastic Net based regularized regression. Next, spatial contextual information is used for refining the classification results obtained in the first step. This is achieved by means of a generic but powerful bilateral filtering post-processing, with a color guidance image retrieved from the principal components of the hyper-spectral image. Under the generalized Elastic Net framework, our proposed model showed the less time complexity. When comparing three widely used hyper-spectral data sets with the other classification methods, our method has shown the noticeable classification accuracy while the number of training samples is relatively small.