A Globally Generalized Emotion Recognition System Involving Different Physiological Signals.

Area of Science:

• Computer Science
• Biomedical Engineering
• Affective Computing

Background:

• Subject-dependent machine learning for human emotion recognition shows high performance but lacks universality.
• Current research is shifting towards subject-independent approaches for broader applications in areas like driver assistance and smart homes.
• Developing universal emotion recognition systems requires models trained on one group and tested on unseen individuals.

Purpose of the Study:

• To explore a novel, robust subject-independent human emotion recognition system.
• To develop a system that can generalize emotion recognition across different individuals and potentially different sensor setups.
• To improve the universality and applicability of machine learning models for emotion detection.

Main Methods:

• The proposed system integrates two core models: an automatic feature calibration model and a classification model.
• The classification model is based on Cellular Neural Networks (CNN).
• The system utilizes physiological signals including Electrocardiogram (ECG), Electrodermal activity (EDA), and Skin-Temperature (ST).

Main Results:

• The system achieved state-of-the-art results with an accuracy rate between 80% and 89% when using identical elicitation materials and sensor brands for training and testing.
• An accuracy rate of 71.05% was obtained when elicitation materials and sensor brands differed between training and testing phases.
• Demonstrated robustness in subject-independent emotion recognition across varied conditions.

Conclusions:

• The developed system offers a robust solution for subject-independent human emotion recognition.
• The approach shows significant promise for real-world applications requiring universal emotion detection.
• Further research can explore broader sensor modalities and diverse subject populations to enhance generalizability.