An approach to EEG-based emotion recognition using combined feature extraction method. (28th October 2016)

• Record Type: Journal Article
• Authors: Zhang, Yong; Ji, Xiaomin; Zhang, Suhua
• Is Part Of: Neuroscience letters. Volume 633(2016)
• Page Start: 152
• Page End: 157
• Publication Date: 2016-10-28
• Subjects: Emotion recognition -- Empirical mode decomposition -- Feature extraction -- Sample entropy -- Support vector machine; Neurology -- Periodicals; Neurology -- Periodicals; Research -- Periodicals; Neurologie -- Périodiques; Neuroanatomie -- Périodiques; Neuropharmacologie -- Périodiques; Neurophysiologie -- Périodiques; Neurology; Periodicals; Electronic journals; 617.48
• DOI: 10.1016/j.neulet.2016.09.037
• Languages: English
• ISSNs: 0304-3940

Abstract

Highlights: Propose an EEG-based emotion recognition method using empirical mode decomposition (EMD) and sample entropy. Only select two channels to calculate IMFs through EMD and use the first 4 IMFs to calculate sample entropies. Analyze the effect of parameters on the results in detail. Experimental results indicate the proposed method is more suitable for emotion recognition than other methods of comparison. Abstract: EEG signal has been widely used in emotion recognition. However, too many channels and extracted features are used in the current EEG-based emotion recognition methods, which lead to the complexity of these methods This paper studies on feature extraction on EEG-based emotion recognition model to overcome those disadvantages, and proposes an emotion recognition method based on empirical mode decomposition (EMD) and sample entropy. The proposed method first employs EMD strategy to decompose EEG signals only containing two channels into a series of intrinsic mode functions (IMFs). The first 4 IMFs are selected to calculate corresponding sample entropies and then to form feature vectors. These vectors are fed into support vector machine classifier for training and testing. The average accuracy of the proposed method is 94.98% for binary-class tasks and the best accuracy achieves 93.20% for the multi-class task on DEAP database, respectively. The results indicate that the proposed method is more suitable for emotion recognition than several methods of comparison.