CAGCNet: generalized contrastive learning for person identification based on channel aggregated EEG features.

Area of Science:

• Biometrics
• Neuroscience
• Machine Learning

Background:

• Person identification using electroencephalogram (EEG) signals, or brainprint recognition, offers high security.
• Existing brainprint recognition methods struggle with distribution differences between training and test data, leading to performance degradation on unseen domain data.
• The large distance between projected features in the latent space exacerbates performance issues in novel environments.

Purpose of the Study:

• To propose a novel channel-aggregated generalized contrastive learning framework for robust EEG-based person identification.
• To address the challenge of performance degradation in unseen domains due to data distribution shifts.
• To enhance the generalization ability of brainprint recognition models.

Main Methods:

• A channel-aggregated generalized contrastive learning framework is proposed.
• Multi-scale convolution with a channel attention block is employed to capture features at different granularities.
• Feature enhancement-based generalized contrast learning augments source domain data at the feature level to improve generalization on unseen data.

Main Results:

• The proposed framework demonstrates superior performance compared to baseline methods on two multi-session datasets.
• The model exhibits enhanced generalization capabilities when applied to unseen domain data.
• Experiments validate the effectiveness of the channel aggregation and generalized contrastive learning approach.

Conclusions:

• The developed channel-aggregated generalized contrastive learning framework significantly improves EEG-based person identification.
• The method effectively mitigates performance degradation caused by unseen domain data.
• This research advances the field of brainprint recognition by offering a more generalizable and secure identification solution.