We’re proud to announce that our work “Trustworthy Super-Resolution of Multispectral Sentinel-2 Imagery With Latent Diffusion” has been officially published in the IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (JSTARS, Vol. 18).
In this paper, we present a latent diffusion-based super-resolution model designed specifically for multispectral Sentinel-2 imagery, enabling spatial enhancement from 10 m to 2.5 m resolution while preserving spectral integrity. Unlike traditional diffusion methods, our model conditions the generation process directly on low-resolution input, ensuring fidelity and minimizing hallucinations. Crucially, it also outputs pixel-level uncertainty maps, providing a powerful trust mechanism for downstream applications like land cover mapping, change detection, and environmental monitoring.
Published under open access, the article highlights the potential of generative models in Earth observation when trustworthiness, spectral consistency, and computational efficiency are prioritized.
🔗 Read the full article here: IEEE Xplore – JSTARS
