Cluster-Wise Weighted NMF for Hyperspectral Images Unmixing with Imbalanced Data

Hyperspectral unmixing is an important technique for analyzing remote sensing images which aims to obtain a collection of endmembers and their corresponding abundances. In recent years, non-negative matrix factorization (NMF) has received extensive attention due its good adaptability mixed data with different degrees. The majority existing NMF-based methods are developed by incorporating additional constraints into the standard NMF based on spectral spatial information hyperspectral images. However, they neglect exploit nature imbalanced pixels included in data, may cause be ignored, thus generally cannot accurately estimated statistical property NMF. To samples during procedure, this paper, cluster-wise weighted (CW-NMF) method proposed. Specifically, result clustering conducted image, we construct weight introduce it model proposed can provide appropriate value reconstruction error between each original pixel reconstructed procedure. way, adverse effect accuracy expected reduced assigning larger values concerning giving smaller endmembers. Besides, extend CW-NMF introducing sparsity abundance graph-based regularization, respectively. experimental results both synthetic real have been reported, effectiveness our demonstrated comparing them several state-of-the-art methods.