Automated accurate emotion recognition system using rhythm-specific deep convolutional neural network technique with...

Area of Science:

• Affective Computing
• Cognitive Neuroscience
• Human-Computer Interaction (HCI)

Background:

• Emotion is a complex psycho-physiological process influencing personality and behavior.
• Recognizing emotions is crucial for advancing human-computer interaction (HCI).
• Electroencephalogram (EEG) signals capture spatiotemporal brain activity relevant to emotion.

Purpose of the Study:

• To develop an automated emotion recognition system using multi-channel EEG signals.
• To propose a rhythm-specific deep convolutional neural network (CNN) approach.
• To classify emotions based on valence, arousal, and dominance dimensions.

Main Methods:

• Utilized multi-channel EEG signals to measure brain electrical activity.
• Applied band-pass filters to extract delta (δ), theta (θ), alpha (α), beta (β), and gamma (γ) rhythms.
• Developed a deep CNN architecture with multiple convolutional, pooling, batch-normalization, and dropout layers for classification.

Main Results:

• The rhythm-specific deep CNN achieved high accuracies (e.g., 98.91% for LV vs. HV using β-rhythm on the DEAP database).
• Specific rhythms (β and θ) demonstrated strong performance in classifying emotional states.
• The model showed varying performance across different databases and rhythm selections (e.g., 57.14% with α-rhythm on DASPS).

Conclusions:

• The proposed multi-channel rhythm-specific deep CNN is effective for automated emotion recognition from EEG.
• The approach holds potential for real-time emotion recognition applications in HCI.
• Further research can explore optimal rhythm selection and model fine-tuning for diverse emotional states.