A manifold learning based feature extraction method with improved discriminative ability

Feature reduction is a key step in hyperspectral image classification. In this paper, we propose a supervised feature extraction method which is based on manifold learning theory. The proposed method uses a new weighting approach in object function to makes between-class samples farther away and makes within-class samples closer in low dimensional feature space. Therefore, discriminative ability of proposed method is improved. The hyperspectral image used in our experiments is collected by AVIRIS sensor over the Indian Pines over a mixed agricultural/forest area. The experimental results show the superiority of proposed method compared to some popular and state-of-the-art feature extraction methods with using limited number of training samples.