A new semi-supervised algorithm for hyperspectral image classification based on spectral unmixing concepts

Spectral unmixing is a fast growing area in hyperspectral image analysis. Many algorithms have been recently developed to retrieve pure spectral components (endmembers) and determine their abundance fractions in mixed pixels, which dominate hyperspectral images. However, possible connections between spectral unmixing concepts and classification algorithms have been rarely investigated. In this work, we propose a new method to perform semi-supervised hyperspectral image classification exploiting the information retrieved with spectral unmixing. The proposed method integrates a well-established discriminative classifier (multinomial logistic regression) with linear spectral unmixing. Furthermore, the proposed method uses a new active sampling approach which takes into account spatial context when generating new samples. The proposed method is experimentally validated using both simulated and real hyperspectral data sets.