SEMI-SUPERVISED DIMENSIONALITY REDUCTION FOR HYPERSPECTRAL REMOTE SENSING IMAGE CLASSIFICATION

Class labels and pairwise constraints are adopted as the prior information to present the semi-supervised dimensionality reduction for hyperspectral image. In this paper, we extend semi-supervised probabilistic principal component analysis (S(2)PPCA), semi-supervised local fisher discriminant analysis (S(2)LFDA) and semi-supervised dimensionality reduction with pairwise constraints (S(2)DRpc) to extract the features of hyperspectral image. These semi-supervised dimensionality reduction approaches are compared with PCA in classification task. Experimental results show that semi-supervised algorithms of S(2)PPCA and S(2)DRpc are superior to PCA.