EEG data augmentation for Personal Identification Using SF-GAN

Because EEG-based identity requires a large amount of training data when training a classification model, and the collection of EEG signals requires a lot of time and effort. Therefore, we hope to perform data augmentation on the EEG data used for identity. Generative adversarial networks have achieved great success in image generation, but the raw EEG signals are not in the form of images. Therefore, we process the EEG signal into an EEG topomap with stronger spatial feature representation, and use a spatial feature-based generative adversarial network image augmentation method (SF-GAN). To verify the generality of our proposed method, we use real EEG topomap samples processed from two different EEG datasets, BCI Competition IV 1 and BCI Competition IV 2a, to train SF-GAN to generate augmented samples for training identity classification model. The proposed method can use smaller real samples to expand the training set of identity, reduce the data dependence on real samples, and reduce the time of data collection to a certain extent. And it is proved by experiments that the data generated by this method can further improve the training effect of the classification model.