Supervised hyperspectral image segmentation using active learning

This paper introduces a new supervised Bayesian approach to hyper-spectral image segmentation. The algorithm mainly consists of two steps: (a) learning, for each class label, the posterior probability distributions, based on a multinomial logistic regression model; (b) segmenting the hyperspectral image, based on the posterior probability distribution of the image of class labels built on the learned pixel-wise class distributions and on a multi-level logistic prior encoding the spatial information. Aiming at reducing the costs of acquiring large training sets, we use active label selection based on the the posterior marginals of the complete model provided by Belief propagation. A comparison of the proposed method with state-of-the-art competitors shows its effectiveness.