Improvised SimCLR Algorithm for Emotion Recognition using EEG Signal

Abstract

The purpose of this endeavor is to create a system that is capable of accurately identifying each emotion that youngsters experience. In order to make children's life better, there is a need for study in the field of interpreting the mental condition of children, which is still developing. On the other hand, voice or image-based emotion recognition algorithms provide false findings due to the fact that children and adolescents will not easily expose their hidden emotions to anyone. The ability to articulate what is going on in their lives can be difficult for some children. People who are affected by this experience depression, which hinders them from living their life as they normally would. EEG waves are utilized to identify children's emotions more precisely than previous SOTA approaches. By doing so, it is possible to determine precisely the state of mind, which is something that cannot be ignored. The findings of this study include convolutional neural network (CNN) models that are merged with Bi-LSTM- SimCLR contrastive learning techniques for the purpose of achieving efficient emotion recognition. The DEAP, which is widely considered to be the most valuable EEG benchmark dataset, is utilized for classification purposes in this study. Additionally, its labels, valence and arousal, dominance and liking, are utilized. In order to represent the dataset, we make use of the Fast Fourier Transformation for the frequency domain features. SimCLR models achieve a flawless accuracy of 97.02% for EEG based image and 95.51% for EEG signal correspondingly, surpassing the performance of every other model that was considered to be state-of-the-art in the past.