Emotion Classification from Short-term EEG Signals in Deep Learning

Emotion is the result of brain processing after the combination of the surrounding environmental state and the individual physiological state. Its essence is the individual perceptual response to the things around him, and the result of the interaction between man and the environment. There is a strong correlation between human cognitive behavior and psychological activities and EEG signals. The physiological activities of the brain can directly reflect individual emotional activities. EEG signal is one of the important input signals in the affective calculation. Correct emotion representation is a key step in emotion recognition research. This paper aims to study emotion EEG recognition through the representation of emotion classification. The author collected 4000 EEG samples (including 64 channels) of four emotions with a time length of 3s, which were neutral, happy, sad, and fear. According to the characteristic that emotions can not be mutated, setting the sample length of 3s can more flexibly grasp the subjects' emotions in a short time. In this paper, two deep learning methods, LSTM and CNN, are used to identify the positive and negative emotions of the new dataset and DEAP (Database for Emotion Analysis using Physical Signal, including 40 channels). The accuracy of the new dataset is 89% and 74% respectively, which are higher than 72% and 70% of DEAP.