Multi-temporal Sentinel-2 super-resolution by optimization
Multi-temporal Sentinel-2 super-resolution by optimization: a method that enhances the spatial resolution of Sentinel-2 satellite imagery by combining multiple images of the same location taken at different times.
Background
- Sentinel-2 is a pair of European Space Agency satellites that capture optical imagery of Earth's surface every 5 days at resolutions of 10m, 20m, and 60m per pixel (the finer the resolution, the fewer spectral bands available). Many downstream tasks — agriculture monitoring, forestry, urban mapping — would benefit from having all bands at 10m.
- "Multi-temporal super-resolution" is a computer-vision technique that fuses multiple, slightly different looks at the same location (taken at different times) to produce a single, sharper image. This is different from the more common single-image super-resolution, which tries to guess detail from just one image.
- This project implements a method that uses an optimization algorithm (rather than a deep neural network) to solve the fusion problem. Given a set of low-resolution 20m and 60m bands acquired over time, it recovers high-resolution 10m bands without training data or a GPU.
- The approach is related to "deep image prior" ideas: the structure of the reconstruction process itself acts as a regularizer, producing plausible high-frequency detail even though no external training dataset is used.