A dual-branch deep learning framework for emotion recognition from EEG signals.

Area of Science:

• Affective Computing
• Biomedical Signal Processing
• Machine Learning for Healthcare

Background:

• Traditional electroencephalogram (EEG)-based emotion recognition faces limitations in manual feature engineering, dataset generalization, and computational complexity.
• These limitations hinder the practical deployment of EEG for real-time mental health monitoring and personalized healthcare.

Purpose of the Study:

• To develop a novel hybrid dual-branch deep learning architecture for robust and efficient EEG-based emotion recognition.
• To minimize preprocessing requirements while enhancing temporal and spectral feature extraction for improved accuracy.

Main Methods:

• A dual-branch deep learning framework integrating Long Short-Term Memory (LSTM) for temporal features and Convolutional Neural Networks (CNNs) for spectral features (Mel-Frequency Cepstral Coefficients - MFCC).
• Incorporation of cross-modality enhancement mechanisms, including inverse MFCC computation and LSTM-to-MFCC projection, for bidirectional feature learning.
• Feature fusion via element-wise multiplication and concatenation, followed by classification using an Artificial Neural Network (ANN).

Main Results:

• Achieved high accuracies on benchmark datasets: 96.49% (Brainwave EEG), 99.99% (WESAD), and 99.99% (SWELL).
• Attained perfect precision, recall, and F1-scores on the WESAD and SWELL datasets, establishing new state-of-the-art performance.
• Demonstrated superior performance over existing methods with preserved computational efficiency suitable for real-time applications.

Conclusions:

• The proposed dual-branch deep learning framework offers a robust and efficient solution for EEG-based emotion recognition.
• The model shows significant potential for practical applications in mental health monitoring and affective computing.
• Availability of source code facilitates further research and development in the field.