Low-Rank Tensor Tucker Decomposition for Hyperspectral Images Super-Resolution

Spatial resolution enhancement method for hyperspectral image based on spatial-spectral feature fusion

“…Therefore, the main trend of HSI spatial resolution enhancement is inspired from image fusion, in which low-spatial resolution HSI and high-spatial resolution images (MSI or PAN) are fused to achieve a recovered HSI with high-spatial resolution by introducing accurate and reliable spatial information. The current HSI spatial resolution enhancement methods based on information fusion can be divided into four categories: The first category is based on pan-sharpening [11][12][13][14][15][16], The pan-sharpening method is briefly described in literature [11]; The second category is based on matrix decomposition methods [17][18][19][20], Yokoya et al [19] proposed a fusion method for coupled non-negative matrix decomposition (CNMF) by alternately unmixing hyperspectral and multispectral data into endmember and abundance matrices; The third category is based on tensor decomposition methods [21][22][23], Jia et al [23] proposed the method for HSI-SR based on low-rank tensor Tucker decomposition and weighted 3D total variance, utilizing the low-rank characteristics and local smoothness prior knowledge of the data; and the fourth category is based on Deep Learning methods [24][25][26][27][28][29], Han et al [30] proposed a clustering-based fusion method that combines HMS and LHS images using multi-branch BP neural networks for nonlinear spectral mapping from HMS to HHS images(CF-BPNNs).…”