Variational Autoencoders For Hyperspectral Unmixing With Endmember Variability

Spectral signatures are usually affected by variations in environmental conditions. Thus, spectral variability is one of the most significant problems in hyperspectral unmixing. Although there have been many spectral unmixing methods to address spectral variability, it is still a non-trivial task to model the endmember variability. This paper presents a variational autoencoder (VAE) framework for hyperspectral unmixing accounting for endmember variability. The endmembers are generated using the posterior distributions of the latent variables to describe their variability in the image. Compared with other existing distribution-based methods, our proposed method is able to fit an arbitrary distribution of endmembers for each material through the representation capacity of deep neural networks. Our proposed method is evaluated using both synthetic and real datasets. The unmixing results show the priority of our proposed method compared with other state-of-the-art unmixing methods.