An emotion recognition method based on frequency-domain features of PPG.

Area of Science:

• Physiological modeling
• Signal processing
• Affective computing

Background:

• Photoplethysmography (PPG) signals contain rich physiological information.
• Extracting frequency-domain features from PPG may offer novel insights into emotional states.
• Current emotion recognition methods can be improved with more robust physiological markers.

Purpose of the Study:

• To systematically analyze frequency-domain components of PPG signals using physiological model simulation.
• To extract key features from PPG signals for emotion recognition.
• To investigate the efficacy of these frequency-domain features in distinguishing emotional states (arousal and valence).

Main Methods:

• Employed a dual windkessel model for PPG signal frequency component analysis.
• Collected concurrent physiological (PPG) and psychological data.
• Utilized support vector machine (SVM) classification and compared with pulse rate variability (PRV) and morphological features.
• Validated findings using the DEAP dataset.

Main Results:

• PPG frequency-domain features significantly differentiated arousal (87.5% accuracy) and valence (81.4% accuracy).
• Validation on the DEAP dataset showed consistent performance (73.5% arousal, 71.5% valence).
• Feature fusion including proposed frequency-domain features improved classification accuracy to over 90%.

Conclusions:

• Physiological modeling of PPG signals effectively extracts features for emotion recognition.
• Frequency-domain PPG features are valuable physiological markers for emotional states.
• This approach provides a foundation for advanced emotion recognition systems.