The Hyperspectral Image Clustering Based on Spatial Information and Spectral Clustering

Abstract

The hyperspectral image (HSI) clustering can extract valuable clustering information and be used to classify the ground truth, monitor environment and so on. Spectral clustering is one of the most popular clustering approaches. The algorithm has been applied in HSI clustering well and has drawn massive attentions. However, most of these methods do not take advantage of spatial information of HSI, which can strengthen the correlation of pixels and increase the accuracy. In this paper, based on the physical characteristics of HSI, a new approach is proposed, called hyperspectral image clustering based on spatial information and spectral clustering (SISC). Combining both spatial window and spectral factor, the algorithm makes use of joint spatial-spectral information, and reconstructs the center point and reveals the local spatial structure by using the spatial nearest points. Experimental results with widely-used hyperspectral data indicate that SISC has a better performance in hyperspectral image clustering.