Integration of Hyperspectral Image Classification and Unmixing for Active Learning

Spectral unmixing is a growing area in remotely sensed hyperspectral image analysis. Many algorithms have been developed to retrieve pure spectral components and deter mine their sub-pixel abundance fractions in this kind of data. 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. Our main contribution is the integration of a well-established discriminative classifier (the multinomial logistic regression) with a state-of-the-art technique for linear spectral unmixing (fully constrained least squares abundance estimation). Furthermore, we propose a new active sampling approach which integrates both the spatial and the spectral information. Our experimental results, conducted with a well-known hyperspectral image data set collected by the Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS) over the Indian Pines region, NW Indiana, reveal that the proposed method can benefit from the newly developed integrated framework.