Weighted Residual NMF with Spatial Regularization for Hyperspectral Unmixing

This paper proposes a weighted residual nonnegative matrix factorization (NMF) with spatial regularization to unmix hyperspectral data. NMF decomposes into the product of two matrices. However, is known be generally sensitive noise, which makes difficult retrieve global minimum underlying objective function. To overcome this limitation, we include weighting mechanism in conventional formulation. strategy treats each row based on factor. In manner, residuals large values are penalized less and small more make unmixing problem robust. Furthermore, weight term form an ℓ1 norm regularizer provide information abundance matrix. Experimental results conducted validate effectiveness proposed method.