Hi, I’m trying to utilized deep learning to analyze spectroscopy data. in the given photo below which belongs to XPS data. so basically imagine each peak represents an element from the periodic table and I have a dataset of all the elements for each element I have 3 samples with different background noise and peak intensity and from those, I augmented the dataset I used for training using Dirichlet distribution. then I tried converting the data to spectrograms and fed it to resnet 16, 34, 50, densenet… and it overfitted, I also tried converting the data to GAF (gramian angular field) it seems to be better but the problem I’m encountering I need to find an architecture or a way to train on elements only and be able to make inference on compound (multiple peaks)
1 Like