چكيده به لاتين
Emotions play a fundamental role in interpersonal communication and in the daily life of humans. Understanding emotions and responding proportionally to them is very important for the quality of life. Emotional intelligence is a types of intelligence that emphasizes the ability to perceive, understand and respond to emotions. The gap between human and machines is bridged by understanding emotions. Therefore, researchers are developing computational models of emotions to reduce distance between humans and machines. In this study, the electroencephalography (EEG) database, SEED IV is used for emotion recognition. The data include 15 healthy subjects recorded during the watching video clips that induce one of the four neutral, sad, fear and happy emotions. Effective connectivity were calculated after preliminary data processing and the extraction of common five-band EEG signals. To estimate effective connections between all frequency bands, transfer entropy was used. Using the mutual information method, 10 features with higher discriminability were selected as the input features to the deep neural network. The highest accuracy belonged to the emotion of sad in the gamma band with 89.9%. The highest accuracy in neutral emotion was 86% in the overall EEG signal (without decompose into bands). The highest accuracy in fear was 85% in the beta band, and for happiness emotion, it was 88% in overall frequency of EEG signal. Moreover, among all examined frequencies, the overall signal accuracy was the highest with a value of 86%. With the increasing development and attention of researchers in the field of emotions and their identification, considering their extensive applications such as brain-computer interface systems, wearable devices, and monitoring mental states for personal use, professional use, and psychotherapy, in this study, using the proposed method and the obtained results, a higher accuracy is achieved compared to the SEED IV database group at Shanghai Jiao Tong University, and their presented article.
