Spectral-Spatial Graph-Based Semi-supervised Hyperspectral Image Classification

Abstract

One of challenges for hyperspectral image classification is small training set versus high dimensional data. Graph-based semi-supervised learning method, which is capable of exploring the intrinsic geometric structure in hyperspectral image data, has been applied to relieve the obstacle of small training set. Hyperspectral images are rich in spectral and spatial information. In this paper, we present a novel spectral-spatial graph-graph semi-supervised hyperspectral classification method. First, SNN-graph and KNN-graph are constructed to capture spatial dependency and spectral similarity, respectively. Then, a PCA process is employed to integrate the two graphs into a spectral-spatial graph. Finally, a semi-supervised learning is performed on this graph. Experiments on real hyperspectral image data demonstrate the effectiveness of our method.