The mind controlled brain computer interfaces would provide new opportunities for people with different disabilities. Furthermore, it could revolutionize business areas, like online networking and game industry. The currently available brain-machine interface systems are slow and inaccurate because of the low signal to noise ratio and the inefficient software. Nowadays the so called deep learning algorithms are reaching growing result in classification problems, like image processing and voice recognition.
The topic of my thesis is to research how deep neural networks can work on EEG signals. The project can be splatted in to three main part:
1. Preparing online available EEG recordings,
2. Test neural networks with different structures on the same dataset,
3. Evaluation of the data.
From the online available datasets, I used the one created by Gerwin Schalk, which contains recordings of 109 subject.
In this work I have realized a sub-module of an EEG-based brain-computer interface algorithm: the developed deep learning neural nets distinguish between the neural activities of open- and closed-eyed subjects. The neural nets can form the basis of more advanced deep learning algorithms, utilized for classification of mental commands, furthermore, they can be used to determine the awareness of the brain-computer interface users.