4D attention-based neural network for EEG emotion recognition

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Electroencephalograph (EEG) emotion recognition is crucial for brain-computer interfaces.
• Current deep learning methods struggle to fully utilize multimodal EEG signal information.
• Advanced feature extraction and attention mechanisms are needed for robust emotion recognition.

Purpose of the Study:

• To propose a novel four-dimensional attention-based neural network (4D-aNN) for enhanced EEG emotion recognition.
• To effectively integrate spatial, spectral, and temporal information from EEG signals.
• To achieve state-of-the-art performance in EEG-based emotion recognition.

Main Methods:

• Raw EEG signals were transformed into 4D spatial-spectral-temporal representations.
• A convolutional neural network (CNN) processed spatial and spectral information with attention.
• A bidirectional Long Short-Term Memory (LSTM) incorporated temporal attention for dependency analysis.

Main Results:

• The 4D-aNN model achieved state-of-the-art performance on DEAP, SEED, and SEED-IV datasets.
• Experimental results validated the effectiveness of attention mechanisms across different domains.
• The model demonstrated superior ability in capturing complex EEG signal characteristics.

Conclusions:

• The proposed 4D-aNN method significantly advances EEG emotion recognition capabilities.
• Attention mechanisms in multiple domains are vital for maximizing information utilization in EEG signals.
• This approach offers a promising direction for future brain-computer interface development.