Emotion Recognition Based On EEG

Emotion is an indispensable part of normal life, a?ecting our workand study. So that it's important to solve the problem of how to recognizeemotion. In this paper, EEG signals are used in the research on emotionrecognition. First, we designed a video experiment to arouse di?erent e-motions of subjects and collected the EEG data. Next we ?ltered thesedata into delta, theta, alpha, beta and gamma frequency bands, then thelog band energy of each electrode and each band were computed as fea-tures. In order to remove the noise unrelated to emotion task, a lineardynamic system approach was applied to smooth the original features, sothat the ?nal features were created. Although individual di?erences do ex-ist, we still try to ?nd the subject-independent features. In this study, we?nd the ?rst 50 subject-independent features most relevant with emotionthrough correlation feature selection method and deeply ?nd the brain areamost relevant with emotion according to the distribution of these features.Principle component analysis algorithm was then used to reduce the fea-ture dimension againon thesetof these subject-independentfeatures. Andan average classi?cation result of 88.5% was obtained with a linear-SVM.Finally, a manifold model was applied to ?nd the trajectory of emotionchanges.