I. Introduction
Hyperspectral image (HSI) is captured by hyperspectral sensors with the high spectral resolution, which can provide both detailed physical properties and spatial distribution of land covers [1]–[4]. A significant task in HSI analysis is HSI classification. It assigns a unique class label to each hyperspectral pixel. Different kinds of feature extraction methods have been developed to utilize plentiful spectral properties and spatial information for accurate HSI classification, such as filtering-based methods and morphology-based methods [5]–[9].