Emotion recognition based on low-cost in-ear EEG

In this paper, we propose a low-cost in-ear EEG device which is implemented by refitting a commercial scalp EEG device, in order to recognize emotion in a manner that is simple, inexpensive, and popular in style. EEG signals of twelve subjects were recorded under three emotion conditions that were induced by music and video materials. By using wavelet packet transformation (WPT), two frequency features and a nonlinear feature are extracted to create a three-dimensional feature vector for each labeled EEG segment. These feature vectors are input into a support vector machine (SVM) classifier for automatic emotion recognition. The SVM classifier achieved a best 94.1% cross-validation accuracy for positive (high valence, HV) and negative (low valence, LV) two-class emotion recognition. However, the accuracy for excited (high valence and high arousal, HVHA), relaxed (high valence and low arousal, HVLA) and negative (LV) multi-class emotion classification was 58.8%. The experimental results show that the proposed low-cost in-ear EEG has outstanding accuracy for valence recognition, but poor accuracy for arousal recognition.