Dissimilarity in Semisupervised Classification of Hyperspectral Image

Hyperspectral image has some typical characteristics such as too many of bands, hard to obtain labeling samples, easy of being interfered of spectral information. These characteristics lead to the dimensionality disaster and the spatial variability of spectral information. To solve these problems, it is proposed to introduce dissimilarity in Laplacian support vector machine (Diss-LapSVM) by the adding dissimilarity information to machine’manifold regularization term , which restrains the influence of the spatial variability effectively. Meanwhile, in order to introduce appropriately distribution of unlabeled samples, This paper provides linear neighborhood propagation (LNP) to construct graph Laplacian matrix. The results illustrated that the proposed method can improve the classification accuracy, especially for samples which have similar spectral features.