The subject of this thesis is blind hyperspectral unmixing using deep learning based autoencoders. Two methods based on autoencoders are proposed and analyzed. Both methods seek to exploit the spatial correlations in the hyperspectral images to improve ...