EEG-Based Emotion Recognition Using Quadratic Time-Frequency Distribution.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Accurate human emotion recognition is crucial for human-machine collaboration.
• Electroencephalogram (EEG) signal analysis for emotion recognition faces challenges due to non-stationarity and feature extraction complexities.

Purpose of the Study:

• To present a novel EEG-based emotion recognition approach utilizing a quadratic time-frequency distribution (QTFD) for enhanced feature extraction.
• To evaluate the efficacy of the proposed method across different emotion labeling schemes and analysis scenarios.

Main Methods:

• Employed QTFD to create high-resolution time-frequency representations of EEG signals.
• Extended 13 time- and frequency-domain features to the joint time-frequency domain for quantification.
• Utilized a 2D arousal-valence plane for four emotion labeling schemes and subject-specific support vector machine classifiers.

Main Results:

• Achieved average classification accuracies ranging from 73.8% to 86.2% across different emotion schemes.
• Demonstrated the effectiveness of the QTFD-based approach in discriminating between various emotion classes.
• Outperformed several existing state-of-the-art EEG-based emotion recognition studies.

Conclusions:

• The proposed QTFD-based method is effective for EEG-based emotion recognition.
• The approach offers a robust way to capture spectral variations for improved emotion classification.
• Further analyses confirmed the method's capability across different brain regions and feature sets.