Sparse Constrained Low Tensor Rank Representation Framework for Hyperspectral Unmixing

Recently, non-negative tensor factorization (NTF) as a very powerful tool has attracted the attention of researchers. It is used in unmixing hyperspectral images (HSI) due to its excellent expression ability without any information loss when describing data. However, most existing methods based on NTF fail fully explore unique properties data, for example, low rank, that exists both spectral and spatial domains. To this low-rank structure, paper we learn different representations HSI spectral, non-local similarity modes. Firstly, divided into many patches, these patches are clustered multiple groups according similarity. Each group can constitute 4-D tensor, including two modes, mode mode, which strong properties. Secondly, regularization with logarithmic function designed embedded framework, simulates spatial, modes tensors. In addition, sparsity abundance also integrated framework improve performance through L2,1 norm. Experiments three real data sets illustrate stability effectiveness our algorithm compared five state-of-the-art methods.