Combining Spectrograms from Different channels of EEG data.

I have EEG data from 25 channels and I wish to combine the STFT spectrograms from those 25 channels into one single image for feeding the data into a CNN. I tried using imagesc to obtain a single spectrogram but when I feed the resulting dataset into the CNN, it gives a very miniscule accuracy, so I wanted to know whether there is some other way of combining the data from the different channels as well.

Answer 1

[Nadia Shaik]

Hi Atharv,

I understand that you want to train CNN using EEG data from 25 channels. You can use a pretrained CNN(for example: Googlenet) by replacing the imageInputLayer, first convolutionLayer to match the input image channel dimension & convolution filter dimensions. You can then fine tune hyperparameters for better accuracy.

Refer to the below link for a similar query.

https://in.mathworks.com/matlabcentral/answers/561122-how-to-do-transfer-learning-with-multi-channel-eeg-data-with-googlenet

I hope it helps.