I. Introduction
Hyperspectral imaging (HSI) technique has made great progress with the development of remote sensing observation systems over the last decade [1], [2]. Taking the advantage of abundant spectral information, HSI is widely applied in many fields, such as land-cover classifications, resource investigation, agricultural planning, and environmental monitoring [3], [4]. The classification for HSI is very important for the above-mentioned applications [5], [6], but the performance of HSI classification usually suffers from the strong spectral correlation affected by the complex distribution of ground objects.