Graph-based semi-supervised hyperspectral image classification using spatial information

Hyperspectral images classification has been one of the most popular research areas in remote sensing community in the past decades. However, there are still some difficulties that need specific attentions, such as the lack of enough labeled samples for training the classifier and the high dimensionality problem, which degrade the supervised classification performance dramatically. The main idea of semisupervised learning is to overcome the contribution of unlabeled samples, which are available in an enormous amount. In this paper, we propose a graph-based semisupervised classification method, using both spectral and spatial information. More specifically, two graphs are constructed and each one exploits the relationship among pixels in spectral and spatial spaces respectively. Then, the Laplacians of both constructed graphs are merged in order to form a weighted joint graph. The experimental results are carried out on Indian Pine AVIRIS image data. The efficiency and the excellent performance of the proposed method is clearly observed in comparison with well-known supervised classification methods, such as SVM, in both terms of accuracy and homogeneity of the produced classified maps.