Hyperspectral Image Classification

Abstract

Hyperspectral image with huge dimensionality is tough to process and classify. To deal these kind of images the standard classifiers like sparse representation and Support vector machine is considered but with slight modifications. SRC, is a linear representation here regularization of classification error, discriminative sparse error, inclusion of contextual information is done. Joint sparse representation is defined. Hyperspectral image is have noise that may be atmospheric noise, Gaussian noise and sparse noise. JSRC is robust to noise and outliers. Support vector machine is supervised classification. Since the Hyperspectral image is high dimensionality, OSP-Orthogonal subspace projection is used here which involves rejection of interference and maximization of Signal to Noise ratio, which reduces noise and dimensionality in one go. SVM may be linear or nonlinear, it depends on type of input data. SVM is more capable when it is having kernel functions. Using SVM we can also classify multiclass data. The classifiers are implemented on two datasets (a) Indian pines data set and (2) University of Pavia and results are displayed.