Two-dimensional empirical wavelet transform based supervised hyperspectral image classification

Hyperspectral image classification is one of the major field of application for hyperspectral imaging systems. Though hyperspectral data gives accurate results than their multispectral counterparts, they are computationally more complex due to their high dimensionality. One of the classical problem while dealing with supervised hyperspectral classification is the class imbalance problem that arises due to the limited availability of samples for training. In order to deal with high dimensionality, many feature mining techniques has been proposed in literature for hyperspectral images. In this paper, we propose a hyperspectral image classification method based on two-dimensional Empirical Wavelet Transform (2D-EWT) feature extraction and compare it with that of Image Empirical Mode Decomposition (IEMD) based extracted features and raw features. Here, the focus is upon the fact that the number of features trained should be less than what is to be tested. Since the computational time for classification is also of prime importance, only some of the fast and best of the classifiers are selected. Sparse-based classifiers are one of the fast and efficient method for supervised classification of hyperspectral images. Subspace Pursuit (SP) and Orthogonal Matching Pursuit (OMP) algorithms are used in our experiments for sparse-based classification. Other classifiers used are Support Vector Machine (SVM) and Hybrid Support Vector Selection and Adaptation (HSVSA). The proposed methodology gives improved performance in terms of classification evaluation measures for hyperspectral image classification task.