| Title |
Optimal Design of Convolutional Neural Network for EEG -based Authentication |
| Authors |
(HyeonBin Lee) ; (Gwangho Kim) ; (JuHyeong Kim) ; (YoungShin Kang) ; (Cheolsoo Park) |
| DOI |
https://doi.org/10.5573/IEIESPC.2021.10.3.199 |
| Keywords |
EEG; Authentication; Bayesian optimization; Deep learning |
| Abstract |
An electroencephalogram (EEG) is an electrical recording from the scalp when neurons in the brain are active. EEG signals have been studied for authentication because they are difficult to falsify and can distinguish individuals. On the other hand, EEG is nonstationary, and its patterns vary slightly. The authentication model was trained day-to-day to overcome the nonstationarity of EEG. EEG signals were measured on two-channel frontal electrodes for five days from 10 subjects in their resting states. Convolutional neural networks were designed for an EEG-based authentication system, and the model was optimized using a Bayesian optimization method. The proposed neural network model was trained with the EEG data from the first to the fourth day and tested using the fifth-day data, which yielded a mean accuracy of 93.23%, precision of 71.31%, and recall of 57.65%. The incremental learning of the EEG signals day-to-day improves the authentication performance, including various EEG patterns in the model. |