The Laser Molecular Imaging and Machine Learning laboratory team published an article titled “Methods for merging hyperspectral and multispectral images based on pixel-by-pixel spectral transformations and cycle consistency” in the journal “Optics and Spectroscopy”. (К2).

Two approaches to fusion of hyperspectral and multispectral images were presented. The first approach is based on the WMatrixGenerator module, which generates pixel-by-pixel spectral transformation matrices with quadratic correction. The second approach utilizes the HSIMSICycleGAN architecture with cycle consistency and physically motivated loss. Experimental evaluation demonstrated the competitiveness of the proposed methods in terms of PSNR, SSIM, SAM, and ERGAS metrics on the Real HSI/MSI/PAN dataset. The methods operate on paired observations of hyperspectral and multispectral images of the same scene but do not require high-resolution reference data.

https://journals.ioffe.ru/articles/63345