Emotion Recognition Based on Weighted Fusion Strategy of Multichannel Physiological Signals.

Area of Science:

• * Affective computing and pattern recognition.
• * Multimodal human-computer interaction and biosignal processing.

Background:

• * Emotion recognition is a key challenge in pattern recognition, utilizing diverse human data like visual, audio, and physiological signals.
• * Existing methods often analyze signals independently, potentially missing synergistic information for accurate emotion detection.

Purpose of the Study:

• * To propose and evaluate a decision-level weight fusion strategy for emotion recognition.
• * To enhance the accuracy of emotion recognition by effectively integrating multichannel physiological signals.

Main Methods:

• * Selection of four physiological signals: Electroencephalography (EEG), Electrocardiogram (ECG), Respiration Amplitude (RA), and Galvanic Skin Response (GSR).
• * Feature extraction across various analysis domains for each physiological signal.
• * Independent classification using Support Vector Machine (SVM) and a feedback strategy for weight definition based on individual signal performance.
• * Decision-level fusion of classifiers using a weight matrix and linear combination.

Main Results:

• * The proposed weight fusion strategy achieved the highest accuracy on the MAHNOB-HCI database.
• * Demonstrated the effectiveness of combining multiple physiological signals through a weighted decision-level approach.

Conclusions:

• * The decision-level weight fusion strategy offers a robust method for multichannel physiological signal-based emotion recognition.
• * Findings support the development of more specialized emotion recognition systems by leveraging weighted fusion of biosignals.