Incremental EEG Biometric Recognition Based on EEG Relation Network

Jianghong Kang, Na Lu, Xu Niu

Abstract: Electroencephalogram (EEG) has drawn increasing attention in biometric recognition field as an emerging high-security biometric. However, personal EEG varies significantly with time and subjects’ mental state, which needs to adjust or add new data and subjects to the original EEG identity database. In such cases, retraining of model leads to high computational cost and transfer learning sometimes does not work well. To address the incremental EEG biometric recognition problem, we proposed a novel method to implement incremental verification and identification using short-time EEG signals of resting state based on a specially designed EEG Relation Network (EEG-RN). The template comparison mechanism in EEG-RN has enabled incremental identification of unseen subjects. In the proposed network, an embedding module is designed to extract discriminant features of inter-subject samples. A relation module computes the similarities between the template samples and the test samples to determine the identification label. Template matching and fine-tuning were employed in the experiments to verify and classify unseen subjects with the model trained on the original dataset. We evaluated EEG-RN on public dataset and the results show that the proposed method can effectively learn efficient features from multiple-subject recognition and improve the practicability of EEG biometric recognition system.