The repository contains two examples of using convolutional neural networks to model spectroscopical data with data augmentation or EMSC as means to handle the variation in baseline.
The first example show what kind of results are possible using data augmentation:
The other example use the same setup, but with EMSC used for baseline correction:
The approach and background + a comparison with PLS models are further described in https://arxiv.org/pdf/1710.01927 and the blog post https://www.wildcardconsulting.dk/useful-information/deep-chemometrics-deep-learning-for-spectroscopy/
Feel free to leave a comment on the blog post if you find it useful ;-)
Please cite: https://arxiv.org/abs/1710.01927
Commercial support discontinued. Wildcard pharmaceutical consulting: https://www.wildcardconsulting.dk/useful-information/wildcard-pharmaceutical-consulting-will-be-closed/