Pattern classification of EEG signals using a log-linearized Gaussian mixture neural network

In this paper, we propose a pattern classification method of EEG (electroencephalogram) signals measured by a simple and handy electroencephalograph to evaluate possibility of the EEG signals as a human interface tool. Subjects are asked to switch their eye states or exposed a flash light turning on and off alternatively according to pseudo-random series for 450 seconds. The EEG signals are measured during experiments and used for classification. Each EEG signal may have different distribution depending on two states of the stimulation such as eye opening/closing and presence/absence of the flash light. Therefore a log-linearized Gaussian mixture neural network incorporated a statistical model is used. It is shown from the experiments that the EEG signals can be classified sufficiently and classification rates change depending on the number of training data and the dimension of feature vectors