I've a set of imagery (0.5m resolution) and coincident 10m RGB Sentinel-2 image for the same AOI. I was wondering if it's possible to use both for increasing the the resolution of another unseen S2 images over the same AOI ?? Ideally to end up having 0.5m Sentinel-2 image.
I have followed the steps explained in the below notebook sample and I assumed that to degrade 0.5m resolution to 10m (to match Sentinel-2), I need to input the downsample factor = 20 (correct me if I'm wrong).
The way it works is that you train the model by giving it a downsampled image (10m) and the same original image at normal resolution (0.5m). The model then works out how to reproduce the higher resolution from the other. So you can't really use the Sentinel imagery in this step. Then you would apply your model to the Sentinel 10m imagery to create an 0.5m Sentinel image.
Now if your training imagery doesn't look like your Sentinel imagery, then your results may not be great. Also, trying to enhance imagery by 20 times might be asking a bit too much - there's simply not enough source imagery to work with. However, here is a page showing Sentinel imagery at 1.5m using a model trained on 1.5m Spot data: Enhancement of Sentinel-2 images at 1.5m
Thanks a lot for your reply. I got your point on how the model works.
I've checked the link you attached, and tried to use t but with no joy as they use otb-applications with isn't starightforward to install. It needs Python 3.5 which is out-of-date and no longer supported by many Python IDEs.
Do you think a decent resolution can be achieved on Sentinel-2 using ArcGIS Python API? What would be the reasonable down-sample factor?