A Dual-Branch Approach to Hyperspectral Image Super-Resolution Using Diffusion Models and Spectral Attention Blocks

Hyperspectral imaging advancements are hindered by the trade-off between spatial and spectral resolution. This paper presents a novel dual-branch approach for HSI superresolution that tackles this issue by separately enhancing spatial and spectral features. The model utilizes a modified diffusion model, enhanced with Multi-Head attention and Squeeze-Excitation Networks, to generate high-quality images. The spectral branch introduces a novel Spectral Attention Block (SAB) with a multi-scale approach for effective spectral feature extraction. The model is further enhanced by global feature extraction and channel attention. It surpasses baseline models across various datasets and metrics, showcasing the potential of diffusion models in HSI superresolution.