EEG Emotion Recognition Based on 3D-CTransNet.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Electroencephalography (EEG) is crucial for brain-computer interfaces and emotion computing.
• Current deep learning models struggle with complex EEG features and long-term dynamic changes.
• Algorithmic and structural constraints limit traditional models.

Purpose of the Study:

• To develop a deep learning model for enhanced EEG-based emotion recognition.
• To address performance degradation in recognizing long EEG signal sequences.
• To improve accuracy and processing speed in emotion classification.

Main Methods:

• Proposed a hybrid Convolutional Neural Network (CNN)-Transformer structure named 3D-CTransNet.
• Utilized 3D data input and electrode position mapping for spatial and temporal feature retention.
• Incorporated self-attention mechanism and parallel processing from Transformer architecture.

Main Results:

• Achieved 97.04% classification accuracy for Valence-Arousal emotion recognition on the DEAP dataset.
• Demonstrated superior performance compared to traditional CNN-LSTM hybrid models.
• Showcased improved recognition accuracy and processing speed due to Transformer integration.

Conclusions:

• 3D-CTransNet effectively recognizes complex features in EEG signals with long-term dynamic changes.
• The hybrid CNN-Transformer architecture overcomes limitations of previous models.
• This model offers a significant advancement for EEG-based emotion recognition in brain-computer interfaces.