| （英） |
In order to achieve high accuracy in emotion recognition based on EEG, it is considered effective to simultaneously learn the interrelationships among temporal, spatial, and frequency information contained in the EEG related to emotion. Therefore, we propose a new emotion identification model that incorporates frequency information into a spiking neural network (SNN) that can learn spatio-temporal information. The input data is encoded by decomposing the EEG into the frequency bands of theta, alpha, beta, and gamma waves, and then taking the product of the correlation coefficient between each band and emotion. As a result, we confirmed that the use of HSIC as a correlation coefficient improved the recognition accuracy for both positive and negative emotions. |