Robust Hyperspectral Unmixing in the Presence of Mixed Noise using Joint Sparsity and Total Variation Regularization

Abstract

Hyperspectral unmixing is a vital process in hyperspectral imaging, aiming to estimate constituent endmembers and their fractional abundances within each pixel. Hyperspectral images are often marred by various noise sources. This research delves into the hyperspectral unmixing challenge in a general scenario that accounts for mixed noise, explicitly addressing Gaussian noise and sparse noise. The unmixing model is tailored to harness the joint sparsity of abundance maps, enabling the robust separation of endmembers and abundances from noise. Additionally, a total-variation-based regularization technique is incorporated to capture the smoothness of abundance maps, further enhancing the accuracy of the unmixing process in noisy hyperspectral images.