I am currently working with a CNN with a physical magnitude dataset. The goal is to downscaling one of them based on the others, but previously I would like to do dimensionality reduction. If I have 10 physical magnitudes, this is, 10 features/variables, in images of 100x100 for every day. How is PCA applied on this? For what I found on the internet, first I have to ravel the images, so I have a (10x10000) tensor and now I should apply PCA on this I think, but what do I with the rest of the days? I mean, if I have data for,lets say, 10 days I end-up having a (10x10x100000) tensor, and I do not know how to apply PCA to this.
In my problem, the channels is what I want to reduce, I want to know what magnitudes are relevant.