This question is regarding neural network EEG classification..

9 views (last 30 days)
I have .mat dataset two class , with three dimension. for example : 62 x 512 x 189 I have processed the signals with spectrogram command and [S,F,T,P] = spectrogram(data,256,224,256,fs). Now I need to feed this output to CNN for classification. I am not sure how to feed the stft in cnn, do i need to follow any specific format. Also I dont have any labels in my dataset. how to input labels in TrainNetwork. Can some one please guide me here. Also Is there a way to do this process using svm ? or Can I save this spectrograms as images and perform classification?
  1 Comment
Star Strider
Star Strider on 28 Nov 2021
I am not certain what the task is or what the EEG signals are. I have used simple nets for EEG classification, however mostly either linear discrikminant and nearest neighbour classification (nearly three decades ago, so not receently, and not with MATLAB). The objective in my research was to use known EEG STFT records at each of the ‘10-20’ electrode locations and with respect to time to classify the ‘unknown’ spectra on the basis of the ‘known’ spectra in order to determine the laboratory cognitive task the subject was doing. (It worked, and we got 66% classification accuracy, highly statistically significant because of the large N, and all the more imporessive because of the relatively primitive linear classifier for such data.)
I always encourage searching PubMed for such information. One study that I found using this search command and that might be of interest is ‘Emotional EEG classification using connectivity features and convolutional neural networks’, Neural Netw 2020 Dec;132:96-107 and while it may not be exactly the desired result, the Similar articles and Cited by sections could lead to something closer to the intended application. (It usually takes several such ‘re-directs’ in my experience in order to get the best informaition, simply because the available lliterature is so vast.)
I’m not posting this as an Answer because it isn’t one. I intend it to to facilitate getting the most relevant information for the desired research application. I hope it does!
Good luck!

Sign in to comment.

Answers (1)

Yomna Genina
Yomna Genina on 7 Dec 2021




Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!