A Sturdy Nonlinear Hyperspectral Unmixing

Hyperspectral unmixing (HSU) is a way to process the prediction of the existing endmembers and the fractional abundances (FA) available in all pixels in the hyperspectral images. However, in a practical scenario, hyperspectral image is frequently corrupted due to many types of noises at the time of acquiring phenomenon such as dead-lines, impulse noise (IN), Gaussian noise (GN), and stripes. This type of complicated noise leads to mitigation in the quality of the acquired HSI, by making them to lose the precision process. To address these issues, this article presents a sturdy nonnegative matrix factorization (S-NMF) with integrated fast dissociable non-local Euclidean median and iterative block coordinate descent algorithm (IFD-NLEM-IBCDA), where FD-NLEM eliminates mixed noise from acquired HSI without degrading the original quality, then IB-CDA is utilized to unmix the HSI by solving S-NMF minimization problem. Furthermore, we also provided a solution for a hyper-parameter that is utilized in IB-CDA during the S-NMF minimization. Extensive simulation results, together on real and synthetic HSI, exhibit the superiority of anticipated unmixing under mixed noise conditions over conventional unmixing algorithms.