Superpixel-based Spatial Weighted Sparse Nonnegative Tensor Factorization Unmixing Algorithm

Hyperspectral unmixing aims to correctly estimate the endmembers and their corresponding abundance fractions in an HSI. Many hyperspectral unmixing methods have been proposed, including the longstanding geometry-based, statistics-based and non-negative matrix factorization (NMF)-based unmixing methods. The traditional NMF-based method expands the three-dimensional hyperspectral data into matrix form and decomposes it into the product of the endmember and the abundance, which causes a certain degree of information loss. The matrix-vector nonnegative tensor factorization algorithm solves this problem well by processing hyperspectral data as a tensor and pioneers a new model based on tensor decomposition. However, such methods still suffer from underutilization of image information and unstable performance at low signal-to-noise ratios (SNR). To solve this problem, we proposed a new superpixel-based spatial weighted sparse nonnegative tensor factorization unmixing model (SupSWNTF), which better exploits the spatial information and improve the sparsity of the solution by adding constraints to the abundance matrix. A series of comparative experimental results on synthetic and real-world data sets show that our algorithm achieves the best unmixing results compared to other state-of-the-art algorithms.