EEG temporal-spatial transformer for person identification.

Area of Science:

• Neuroscience
• Biometrics
• Artificial Intelligence

Background:

• Electroencephalogram (EEG) identity recognition offers enhanced security due to signal resistance to theft.
• Current EEG identification methods often rely on single brain states and repetitive stimuli, limiting real-world applicability.
• Human cognitive and emotional states are dynamic, necessitating robust identification techniques adaptable to varying conditions.

Purpose of the Study:

• To develop and evaluate a transformer-based approach for robust EEG person identification.
• To assess the method's generalization capabilities across diverse human states.
• To achieve state-of-the-art performance in EEG biometrics without manual feature engineering.

Main Methods:

• A transformer architecture was employed to process EEG signals.
• Self-attention mechanisms were utilized to extract features from both temporal and spatial domains.
• The proposed method was rigorously tested for its ability to generalize across different cognitive and emotional states.

Main Results:

• The transformer-based method demonstrated superior performance in EEG identity recognition.
• The approach achieved state-of-the-art results compared to existing advanced EEG biometrics techniques.
• The model effectively identified individuals across various brain states, showcasing strong generalization.

Conclusions:

• The proposed transformer model offers a powerful and adaptable solution for EEG-based person identification.
• This method overcomes the limitations of previous approaches by handling diverse and changing human states.
• The elimination of manual feature extraction simplifies the process and enhances practical utility in real-world biometric systems.