It is an interesting article but your data is not balanced. You have different frequency of the different classes. I would suggest you do something about it before proceeding with augmentation. Although, I am not very well aware but you can check if augmentation can be useful to resample your data.
