Hyperspectral Unmixing for Highly Correlated Endmembers using Scaled Endmembers and Abundance Sparsity Constraint NMF

Hyperspectral unmixing (HU) is an essential process in remote sensing that involves decomposing hyperspectral images into distinct spectral signatures (endmembers) and their corresponding fractional abundances. However, handling highly correlated data in HU can be difficult, as traditional methods often perform inadequately. This paper introduces a novel method for hyperspectral unmixing called Scaled Endmembers and Abundance Sparsity Constraint Non-negative Matrix Factorization (SEA-NMF). The method incorporates a scaling mechanism for endmembers to enhance distinctness among the endmembers. Additionally, we include an abundance sparsity constraint to account for the spatial distribution of the hyperspectral image, which in real scenarios has a minimal number of endmembers that contribute to each pixel. Experiments on both synthetic and real hyperspectral datasets have demonstrated that SEA-NMF outperformed existing unmixing techniques in terms of reconstruction error and endmember estimation accuracy, when highly correlated endmembers are involved. The results demonstrate that our method successfully addresses the challenges caused by highly correlated data, offering a robust tool for better hyperspectral image analysis.