Multi-channel EEG emotion recognition through residual graph attention neural network

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electroencephalography (EEG) is a non-invasive technique for measuring brain activity.
• Accurate emotion recognition from EEG signals remains a challenge due to complex neural patterns.
• Existing methods often struggle to fully leverage the spatial and relational information within EEG data.

Purpose of the Study:

• To propose a novel EEG emotion recognition method using a residual graph attention neural network.
• To effectively integrate spatial information from electrode positions and channel connectivity.
• To enhance the accuracy of multi-channel EEG-based emotion recognition.

Main Methods:

• Constructed a 3D sparse feature matrix capturing electrode spatial relationships.
• Utilized a residual network to extract high-level abstract features with spatial context.
• Modeled multi-channel EEG time-domain features using graph theory and a graph attention network.
• Fused features from both networks for final emotion state classification.

Main Results:

• Demonstrated that spatial domain information and inter-channel relationships are crucial for emotion recognition.
• The proposed model effectively fuses these distinct information sources.
• Achieved improved performance in multi-channel EEG emotion recognition on the DEAP dataset.

Conclusions:

• The integration of spatial and relational EEG features significantly enhances emotion recognition.
• The residual graph attention neural network provides an effective framework for this integration.
• This approach holds promise for advancing brain-computer interfaces and affective computing.