I am trying to train a Convolutional Neural Network (CNN) model on an EEG measurement dataset consisting of 32 channels (i.e. 32 non-stationary signals at each time frame/window). In terms of frequency domain analysis, the normal FFT coefficients didn't work out that good so now I'm trying to make a CNN model and use CWT's resulting coefficients as an input.
However, after a bit of googling I figured out you normalise the normal image data before feeding it into a CNN network.
My question is this:
How would you normalise the absolute value of CWT coefficients? From my understanding at each scaling factor the wavelet is shifted along the signal and is integrated at each time, giving back coefficients representing the signal, so does this mean that there is no maximum for a CWT coefficient?