Hyperspectral Image Super-Resolution

Hyperspectral image super-resolution aims to create sharper images without losing the wavelength-by-wavelength information that helps distinguish materials and identify biological or environmental features. Common approaches fuse a spatially detailed RGB, multispectral, or panchromatic image with a lower-resolution hyperspectral image, using CNNs, Transformers, state-space models, or spectral unmixing to combine their complementary information. Other methods use sensor physics and self-supervision, or continuous implicit and operator representations, to work with missing high-resolution targets, unknown sensor responses, or arbitrary output scales. A central challenge is balancing spatial sharpness against spectral accuracy, especially when input images are misaligned or sensor degradations differ; performance can also depend strongly on evaluation conditions and generalization to unseen sensors.