Emotion recognition with reduced channels using CWT based EEG feature representation and a CNN classifier.

Area of Science:

• Neuroscience and Computational Intelligence
• Biomedical Signal Processing

Background:

• Emotion recognition from electroencephalogram (EEG) data is a long-standing research area.
• Existing methods often require extensive computation and lack efficient feature domains for improved classification.
• A need exists for more accurate and computationally less intensive EEG-based emotion recognition techniques.

Purpose of the Study:

• To propose a novel continuous wavelet transform (CWT)-based feature representation for multi-channel EEG data.
• To enhance automatic emotion recognition accuracy while reducing computational complexity.
• To introduce effective channel and CWT scale selection schemes for EEG emotion recognition.

Main Methods:

• Utilized continuous wavelet transform (CWT) to preserve time-frequency information in EEG data.
• Mapped CWT coefficients to a strength-to-entropy component ratio, creating a 2D representation (CEF2D).
• Employed a deep convolutional neural network architecture fed with the CEF2D feature matrix.
• Developed CWT domain energy-to-entropy ratio-based schemes for channel and scale selection.

Main Results:

• Achieved improved classification accuracy for both 2-class and 3-class emotion recognition tasks.
• Reported average accuracies of 98.83% (valence) and 98.95% (arousal) for the 2-class problem.
• Achieved average accuracies of 98.25% (valence) and 98.68% (arousal) for the 3-class problem.

Conclusions:

• Entropy-based features derived from EEG data in the CWT domain are highly effective for emotion recognition.
• The proposed feature domain significantly enhances classification performance compared to previous studies.
• Effective channel selection based on the CWT domain energy-to-entropy ratio effectively reduces computational complexity.