I have an image which has 48 bands. I used PCA then band number is reduced to 12 for representing %99.9 data. But png, jpg hold only 3 bands. How can I use hyperspectral images with Nvidia digits ( Caffe )?
I tried to combine 4 pictures with 3 bands. But the result is very bad. I do not have a problem when I make the Arff file and give it to WEKA. But I could not figure out how to use hyperspectral images with neural networks.
What is the best way of using hyperspectral images with neural networks (Nvidia digits ( Caffe ))?