Motor imagery brain-computer interface powered by artificial neural networks

Abstract

This doctoral thesis focuses on developing a Brain-Computer Interface based on motor imagery Electroencephalogram (EEG) signals using EMOTIV EPOC+ equipment, a SoCKit FPGA development card, and a walking robot. Brain-Computer Interfaces (BCIs) meaningfully improve what was already known as assistive devices for people with disabilities, especially in the lack of global or partial motor skills, employing technological advancements. These brain-computer interfaces enable effective communication between the brain and a given machine using specific cerebrum signals, highlighting challenges such as instant and efficient signal processing, accurate signal decoding and classification, and the conception of universal BCIs using adaptive processing algorithms for all brain signal types. Therefore, EMOTIV EPOC+ headset detects the neuronal activity generated by the defined task and wirelessly sends the corresponding signals to the SoCKit FPGA board for parallel processing using neural networks. Movement imagery signals of right and left fists are processed and converted into operational commands to move the hexapod robot forward or backward. Motor imagery (MI)-EEG signals from the F3, F4, FC5, and FC6 channels are processed using a hybrid architecture of Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. This method uses the deep learning recognition model to develop a real-time embedded BCI system, where signal processing must be seamless and precise. In addition, to deal with the noisy and the non-stationary EEG signal processing problems, two approaches based on the Empirical Mode Decomposition (EMD) method are analyzed. The validation of the results found using the k-fold cross-validation method and two public databases showed the successful functioning of the developed BCI. The bases established in this thesis serve to develop more complex and precise BCIs.