Model-Based Deep Autoencoder Networks for Nonlinear Hyperspectral Unmixing

Autoencoder (AEC) networks have recently emerged as a promising approach to perform unsupervised hyperspectral unmixing (HU) by associating the latent representations with abundances, decoder mixing model and encoder its inverse. AECs are especially appealing for nonlinear HU since they lead model-free algorithms. However, existing approaches fail explore fact that should invert process, which might reduce their robustness. In this paper, we propose model-based AEC considering fluctuation over linear mixture. Differently from previous works, show restriction naturally imposes particular structure both networks. This introduces prior information in without reducing flexibility of model. Simulations synthetic real data indicate proposed strategy improves HU.