Study of Recurrent Neural Networks Applied to Brain-Computer Interface Systems
Brain-Computer Interface (BCI) systems aim to establish a connection from brain signals to devices in the external world and open up possibilities of use in medical applications or even in the field of entertainment. Brain signals obtained by electroencephalograms (EEG) need to be classified, which is a fundamental challenge in BCI systems. EEG signals can be seen as a set of time series, which can be efficiently processed by Artificial Neural Networks, especially the so-called Recurrent Neural Networks (RNNs). In this context, the present research focuses on the use of RNNs, in particular the LSTM (Long Short-Term Memory) network, to process EEG signals on two approaches. The first one considers the modeling of EEG signals through predictive RNNs, whose adjusted weights must represent, in a low dimension, the relevant features for the classification task. By means of comparison with the previous approach, the second approach aims to classify brain signals through architectures already proposed in the literature, such as EEGNet, however modifying it and adding one or more layers with recurrence, in order to improve information processing between the convolutional layers in EEGNet. Through these two approaches, it is expected to obtain a more efficient BCI system for EEG signal processing.