The goal of this thesis was to create a system which moves a cursor on a computer monitor in four directions, based on electroencephalogram (EEG) recordings. The EEG recordings were recorded while imagining of motor activities.
This thesis documents three different protocols for processing EEG records and training a neural network. The purpose of the first protocol is to search for simple relations between amplitude spectra of μ and β brainwaves and imaginary motor activities of opening and closing individual fists. Using the amplitude spectra of μ and β, the neural network achieved 52% accuracy. In the second protocol only the average of amplitude spectra of μ and β was used. Additionally the training set was optimized, resulting in 80% accuracy. The final solution used two neural networks, first one being identical to the one in second protocol and the second one being used to classify imaginary motor activities of opening and closing both fists individually, achieving 69% accuracy.
|
