ARTICLE AD BOX
How do I downscale my Sentinel-2 satellite imagery from the standard 10‑meter resolution to 5 meters. I've looked into the various algorithms and machine learning techniques commonly used for this, but they typically require training models—a time‑intensive process. In a recent meeting with a company that offers 1‑meter resolution services, they mentioned it took them three to four years and hundreds of gigabytes of imagery to train their model. Given my project deadlines, I simply don’t have that kind of time.
Additionally, their pricing is prohibitive, especially since I need to process three years’ worth of image data to support subsequent algorithms. I'm aiming for the most cost‑effective way to perform super‑resolution on these images, nor do I have the hardware to power this process day and night
I’ve come across tools like Real‑ESRGAN, TensorFlow, and Sen2Res, but so far, they haven’t produced results that meet my expectations or satisfy my superiors. My current workflow, which connects to the Copernicus API, is built in C#. Ideally, I’d like to find packages that integrate smoothly within that environment. I’m open to any suggestions or documentation. Right now, I’m already retrieving the bands and the TCI image, running Sen2Cor for atmospheric correction, and then just zoom into the picture with the most amount of detailed and clarity I can get with the image.
