Semi-supervised contextual classification and unmixing of hyperspectral data based on mixture distributions

This paper considers image unmixing of hyperspectral data with a small training data set. We propose a semi-supervised contextual unmixing method for hyperspectral data. Gaussian mixture models and a novel MRF (Markov random field) are assumed for distributions of feature vectors and category fraction vectors, respectively. Then, we derive a semi-supervised unmixing method through EM algorithm and ICM method. The proposed method is examined through artificial and real data sets, and shows a excellent performance.