Hyperspectral image classification with hypergraph modelling

Hyperspectral image classification requires a classifier which can deal with high-dimensional hyperspectral data. How to explore the relationship among different pixels in the hyperspetral image is essential for hyperspetral image classification. In this paper, we propose to formulate the correlation among pixels by using a hypergraph structure. The hypergraph is constructed by using the neighborhood clustering method, where each pixel is connected to its several neighbor pixels. We investigate the relevance scores among these pixels with a hypergraph learning procedure, and hyperspectral image classification is conducted by using these relevance scores. We conduct experiments on the Salinas scene dataset, and experimental results demonstrate that the proposed method can outperform the state-of-the-art methods.