Emotion recognition system based on electroencephalography

A system based on electroencephalography was designed and implemented to analyze the relation of the electroencephalography with the self-assessment manikin evaluation when sound stimuli that provoke emotions are applied. The analysis was made in volunteers between 18 and 40 years old, who were not deaf, to whom a protocol was applied that allowed them to evoke their emotions using sounds. The sounds used during the test were pre-sorted by each volunteer into four main groups, according to the emotional characteristics of activation and valence that correspond to the quadrats presented in the Russell model. A 7-channel electroencephalographic acquisition equipment was used to get the signals from positions T3, C3, F3, FP1, T4, C4 and F4 following the international system 10-20. Then a feature extraction was carried out on the signals for an emotion quadrant classification. Statistical difference was found in rations delta/beta of C3, C4, T3 and T4 positions. It was found that the best classification obtained was with neural networks that achieved an efficiency from 75% to 100%, depending on the participant, for the differences between F3 and F4 of the power spectral density, using the Yule Walker method, and the energy obtained with the Fourier Transform.