Transformer-Based Emotion Recognition with EEG.

Area of Science:

• Biomedical Signal Processing
• Neuroscience
• Artificial Intelligence

Background:

• Emotion recognition from electroencephalography (EEG) is crucial for understanding affective states.
• Traditional methods often process single EEG channels, limiting spatiotemporal analysis.
• Accurate decoding of neural correlates of emotions is vital for advanced applications.

Purpose of the Study:

• To develop a novel Transformer-based model for predicting valence and arousal from multi-channel EEG signals.
• To enhance the analysis of spatiotemporal dynamics in EEG for improved emotion recognition.
• To validate the model's performance on a public dataset for subject-specific emotion decoding.

Main Methods:

• Utilized a tailored Transformer architecture designed to process multiple EEG channels concurrently.
• Implemented a subject-specific training approach for personalized emotion recognition.
• Validated the model using 10-fold cross-validation on the DEAP dataset.

Main Results:

• Achieved high prediction accuracies: 92.66% for valence and 91.17% for arousal.
• Demonstrated the model's capability to capture complex spatiotemporal patterns in EEG.
• Successfully performed subject-specific emotion level prediction.

Conclusions:

• The novel Transformer model offers a robust and accurate method for emotion recognition via EEG.
• This approach enhances the understanding of neural correlates of emotions.
• Potential applications include improved brain-computer interfaces and human-robot interaction systems.