Gaussian mixture model and Markov random fields for hyperspectral image classification

Abstract

This paper presents a novel method for reliable and efficient spatial-spectral classification of hyperspectral data. This algorithm is based on the Bayesian labelling by combining the results of the Gaussian mixture model (GMM) with spatial-contextual information extracted by Markov random fields (MRF). Moreover, a new fuzzy segmentation-based function was defined and incorporated into the spatial energy involved to improve the performance of MRF. To evaluate the proposed algorithm in real analysis scenarios, three benchmark hyperspectral datasets, i.e. Indian Pines, Pavia University and Salinas, were used. Experimental results demonstrated that the proposed method could considerably improve the classification’s overall accuracies when compared to conventional MRF-based approaches.

Keywords

• Hyperspectral image classification
• Gaussian mixture model
• Markov random field