Hyperspectral super-resolution via coupled tensor ring factorization

• Published in: Pattern recognition Vol. 122; p. 108280
• Main Authors: He, Wei, Chen, Yong, Yokoya, Naoto, Li, Chao, Zhao, Qibin
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2022
• Subjects: Coupled tensor ring decomposition; Hyperspectral; Multispectral; Super-resolution; Super-resolution; Multispectral; Coupled tensor ring decomposition; Hyperspectral
• ISSN: 0031-3203; 1873-5142
• DOI: 10.1016/j.patcog.2021.108280

•Based on TR factorization, we developed a degradation model from the HR-HSI to the MSI and HSI. We proposed a CTRF model for HSR tasks. The nuclear norm regularization of the third TR core with mode-2 unfolding was introduced to further exploit the global spectral low-rank property of the HR-HSI.•We analyzed the superiority of the CTRF model for HSR and developed an efficient alternating iteration method for the proposed model. The experiments demonstrated the advantage of the CTRF model compared to the previous matrix/tensor and deep learning methods. Hyperspectral super-resolution (HSR) fuses a low-resolution hyperspectral image (HSI) and a high-resolution multispectral image (MSI) to obtain a high-resolution HSI (HR-HSI). In this paper, we propose a new model called coupled tensor ring factorization (CTRF) for HSR. The proposed CTRF approach simultaneously learns the tensor ring core tensors of the HR-HSI from a pair of HSI and MSI. The CTRF model can separately exploit the low-rank property of each class (Section 3.3), which has not been explored in previous coupled tensor models. Meanwhile, the model inherits the simple representation of coupled matrix/canonical polyadic factorization and flexible low-rank exploration of coupled Tucker factorization. We further introduce spectral nuclear norm regularization to explore the global spectral low-rank property. The experiments demonstrated the advantage of the proposed nuclear norm regularized CTRF model compared to previous matrix/tensor and deep learning methods.