Multiple Convolutional Neural Networks in EEG Emotion Recognition

Abstract

Emotion is a psychological activity in controlling feelings, whether consciously or not. Emotions recognition can use neuropsychology signals obtained through the Electroencephalogram (EEG) device. EEG recording has a multi-channel that is taken from some area in the brain. Each channel provides information from a specific part of the brain, so the signals need to be processed in parallel. The multi-channel recording enriches emotional information so that the signal recognition of each channel is combined with the fusion function. In this way, multi-channel processing does not interfere with the signal sequencing within each channel. Convolutional Neural Networks (CNN) is one of the methods that can learn and recognize patterns in one area, such as the eye. This paper proposed multiple CNN to recognize emotion. First, the EEG is filtered using a wavelet to get a frequency component of 4-45 Hz that represents the characteristics of negative, neutral, or positive emotions. The frequency band contains Theta, Alpha, Beta, and Gamma waves. The experiment gave that Multiple CNN increased accuracy from 64.14% to 80.66% compared to Single CNN. In addition, a wavelet filter to maintain the signal sequence can obtain slightly better accuracy results than wavelet extraction.