Hyperspectral Image Super-Resolution with Self-Supervised Spectral-Spatial Residual Network

Recently, many convolutional networks have been built to fuse a low spatial resolution (LR) hyperspectral image (HSI) and high (HR) multispectral (MSI) obtain HR HSIs. However, most deep learning-based methods are supervised methods, which require sufficient HSIs for training. Collecting plenty of is laborious time-consuming. In this paper, self-supervised spectral-spatial residual network (SSRN) proposed alleviate dependence on mass SSRN, the fusion MSIs LR considered pixel-wise spectral mapping problem. Firstly, paper assumes that between can be approximated by (derived from MSIs) Secondly, explored SSRN. Finally, fine-tuning strategy transfer learned generate SSRN does not as information in Simulated real databases utilized verify performance