CNN and LSTM-Based Emotion Charting Using Physiological Signals.

Area of Science:

• Affective computing
• Biomedical signal processing
• Machine learning for emotion recognition

Background:

• Affective computing relies on physiological signals (EEG, ECG, GSR) for emotion recognition.
• Real-world emotion detection faces challenges in performance and environmental constraints.
• Existing methods struggle with a broad range of emotion classes in unconstrained settings.

Purpose of the Study:

• To propose a subject-independent computational framework for enhanced emotion recognition.
• To utilize deep learning architectures (CNN, LSTM) for processing multi-modal physiological signals.
• To validate the framework on publicly available datasets (DREAMER, AMIGOS) using wearable sensors.

Main Methods:

• A 2D CNN architecture for 14-channel EEG data.
• A combined LSTM and 1D-CNN architecture for ECG and GSR signals.
• Subject-independent analysis using DREAMER and AMIGOS datasets.

Main Results:

• Outperformed state-of-the-art approaches in classifying four emotion classes (valence-arousal).
• Achieved high accuracy with individual modalities: ECG (98.73%), EEG (76.65%), GSR (63.67%) on AMIGOS.
• Multi-modal fusion yielded the highest accuracy: 99.0% (AMIGOS) and 90.8% (DREAMER).

Conclusions:

• The proposed deep learning framework effectively extracts features for emotion recognition.
• Multi-modal fusion of physiological signals significantly enhances emotion elicitation performance.
• The method is suitable for broader applications in less constrained, real-world environments.