I. Introduction
Hyperspectral image data consist of hundreds of observed contiguous wavelength bands that include high-resolution spectral information about various materials in the scene [1]. Thus, the rich spectral information potentially provides information useful for image classification and target recognition. Hyperspectral image classification, which aims at categorizing pixels into one of several land-use land-cover classes, has been raised in many practical remote-sensing applications in recent years, such as environmental monitoring, crop analysis, pollution detections, and abundance estimation, etc. [2]– [4]. However, the relevant challenge should be considered in the classification of the hyperspectral image data with improved accuracy and robustness.