Data Augmentation for Training Deep Neural Networks in EEG Signal Classification
This work focuses on evaluating the application of data augmentation techniques in electroencephalographic (EEG) signals applied to artificial neural networks. The main objective is to evaluate the hypothesis that data augmentation techniques can highlight relevant features in data when applied to networks with convolutional structures such as EEGNet, ShallowNet and DeepConvNet. The use of augmentation techniques to expand the capacity of networks for extracting the main characteristics that define each class allows a broader application of the use of brain-computer interface systems, in such a way that the networks used can make better use of the data available for training. Furthermore, this work proposes the adaptation of a data augmentation method used in convolutional networks for image processing in the context of EEG data. Finally, an analysis of the results obtained is carried out together with the behavior of the networks in each base used.