Emotion Recognition from Physiological Signals Using Parallel Stacked Autoencoders

• Published in: Neurophysiology (New York) Vol. 50; no. 6; pp. 428 - 435
• Main Authors: Bagherzadeh, S., Maghooli, K., Farhadi, J., Zangeneh Soroush, M.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.11.2018; Springer; Springer Nature B.V
• Subjects: Aircraft; Arousal; Biomedical and Life Sciences; Biomedicine; Classification; Deep learning; EEG; Electroencephalography; Electromyography; Emotions; Neurosciences; Physiological aspects; Physiology; deep learning; parallel stacked autoencoder; emotions; central and peripheral signals; electroencephalogram (EEG)
• ISSN: 0090-2977; 1573-9007
• DOI: 10.1007/s11062-019-09775-y

The problem of recognition of emotions in humans is still open from a few aspects. In our study, we used a deep learning method named stacked autoencoder to classify regions of four emotional states in valencearousal plane (high valence-low arousal, high valence-high arousal, low valence-low arousal, and low valence-high arousal). We used a number of physiological signals, including electroencephalogram (EEG), electromyogram (EMG), and other peripheral signals from the DEAP database and extracted spectral and time features from these signals. Also, nonlinear features were extracted from EEG. Then these features were imported to multiple stacked autoencoders in a parallel form (PSAE) to primarily classify four emotional regions in a valence-arousal plane. The final decision about classification was performed using the majority voting method. The average accuracy for classifying four emotional regions from the valence arousal plane, i.e., low arousal and low valence (LALV), low arousal and high valence (LAHV), high arousal and high valence (HAHV), and high arousal and low valence (HALV), reached 93.6%. This result shows that our proposed method demonstrates certain advantages in solving the classification problem; probably, it can also be used for other classification problems.