SPECTRAL GROUPING DRIVEN HYPERSPECTRAL SUPER-RESOLUTION

Convolutional neural networks have been shown to be particularly powerful in extracting low-level concepts in an image. Given the exceptional performance of transformers in exploiting the long-range correlations from an image, many methods are being explored that take advantage of both architectures. To strengthen our network, we add an important feature to transformers using band grouping and a simple CNN architecture to achieve single-image super-resolution (SISR) in hyperspectral imaging (HSI). The primary goal of this work is to train a set of simple residual modelling architectures and then integrate them into a transformer architecture to solve the super-resolution problem in HSI. Further we analyse how swinIR can be adapted to take full advantage of the band-grouping derived information for efficient SISR. Moreover, the proposed architecture gives primary results on standard datasets.