Emotion recognition based on multi-modal physiological signals and transfer learning.

Area of Science:

• Physiological computing
• Machine learning
• Affective computing

Background:

• Emotion recognition from physiological signals faces challenges due to individual differences and data noise.
• Collecting sufficient labeled data for single-subject training is costly and time-consuming.

Purpose of the Study:

• To develop a transfer learning algorithm that overcomes individual differences and noise in physiological signals for improved emotion recognition.
• To enhance the accuracy and efficiency of emotion recognition systems.

Main Methods:

• Proposed a joint probability domain adaptation with bi-projection matrix (JPDA-BPM) algorithm to align feature distributions across subjects.
• Introduced a substructure-based joint probability domain adaptation (SSJPDA) to mitigate noise effects in physiological signals.
• Validated the SSJPDA-BPM algorithm on the DEAP dataset for emotion analysis.

Main Results:

• The SSJPDA-BPM algorithm achieved average recognition accuracies of 63.6% for valence and 64.4% for arousal on multimodal DEAP dataset data.
• Demonstrated significant improvements over the standard joint probability domain adaptation (JPDA) method.
• Showcased a 17.6% increase in valence and a 13.4% increase in arousal recognition accuracy compared to JPDA.

Conclusions:

• The proposed SSJPDA-BPM algorithm effectively addresses individual differences and noise in physiological signals for emotion recognition.
• This transfer learning approach offers a promising solution for developing robust and accurate emotion recognition systems.
• The method significantly enhances performance compared to existing domain adaptation techniques.