A Study On Stable Patterns Of EEG Signals For Emotion Recognition

With the development of artificial intelligence,affective computing has become a popular topic for human computer interaction.Emotion recognition based on computer systems is the key to realizing emotional intelligence.Electroencephalogram(EEG),as one of the most effective physiological signals reflecting emotions,has been recently applied to emotion recognition.This study focuses on the stable patterns of EEG changing with different emotional states.We use video clips as stimuli to invoke people's emotions and collect their EEG signals.We extract differential entropy as EEG features,and use linear dynamical system to do feature smoothing.After that,discriminative graph regularized extreme learning machine and support vector machine are used to classify emotional states from EEG features.Results of this study indicate that differential entropy features can not only express emotions of positive,neutral and negative states,but can also express emotions of happiness,calm,sadness and fear.By calculating the correlation coefficients of EEG features and the emotional states,we get the key brain areas and key frequency bands of emotions,which could provide some evidence for the physiological mechanism of emotion.By using the fuzzy integral algorithm for classifier combination,EEG models can be optimized and better classification results are obtained.Furthermore,we found that for the same subject,the pattern that EEG signals change with emotional states is of time stability.Also for different subjects,their EEG patterns of emotion are similar to some extent.For positive,neutral and negative emotions,there are gender differences in EEG signals,which means EEG patterns from the same gender are more similar.As a result,the stable EEG patterns of emotion make it possible to apply the EEG-based emotion recognition in our daily life.