Related Experiment Video
Updated: Jun 7, 2025
12:48
Investigating Social Cognition in Infants and Adults Using Dense Array Electroencephalography dEEG
Published on: June 27, 2011
17.8K
Affective EEG-based cross-session person identification using hierarchical graph embedding
Honggang Liu1,2, Xuanyu Jin1,2, Dongjun Liu1,2
1School of Computer Science, Hangzhou Dianzi University, Hangzhou, 310018 Zhejiang China.
Cognitive Neurodynamics
|November 18, 2024
Summary
This study introduces a new Multi-scale Convolution and Graph Pooling network (MCGP) for more accurate electroencephalogram (EEG) biometrics. The MCGP model effectively identifies individuals despite emotional state changes, enhancing confidential person identification.
Area of Science:
- Neuroscience
- Biometrics
- Machine Learning
Background:
- Electroencephalogram (EEG) signals are explored for confidential person identification.
- Variations in affective (emotional) states pose a significant challenge to EEG-based biometrics due to the non-stationary nature of EEG signals.
- Accurate EEG-based person identification requires methods robust to affective state fluctuations.
Purpose of the Study:
- To develop and evaluate a novel deep learning network, the Multi-scale Convolution and Graph Pooling network (MCGP), for robust EEG-based person identification.
- To mitigate the impact of affective state variations on the accuracy of EEG biometrics.
- To assess the performance of MCGP across different datasets and experimental conditions, including cross-session identification with mixed and single affective states.
Main Methods:
- An integrated Multi-scale Convolution and Graph Pooling network (MCGP) was employed.
- The MCGP network utilized multiple 1D convolutions at different scales for dynamic feature extraction and fusion.
- A graph pooling layer with an attention mechanism was incorporated to generate hierarchical graph embeddings, which were then fed into a fully connected classification layer.
Main Results:
- MCGP achieved high average accuracies: 85.51% on SEED and 88.69% on SEED-V datasets in cross-session conditions with mixed affective states.
- In single affective state cross-session scenarios, MCGP attained 85.75% (SEED) and 88.06% (SEED-V) for same affective states, and 79.57% (SEED) and 84.52% (SEED-V) for different affective states.
- Results demonstrated that MCGP significantly mitigated the impact of affective state variations compared to baseline methods.
Conclusions:
- The proposed MCGP network effectively addresses the challenge of affective state variations in EEG-based person identification.
- MCGP offers a promising approach for enhancing the confidentiality and accuracy of biometric systems utilizing EEG signals.
- Performance was slightly better when identifying individuals within the same affective state compared to across different affective states in cross-session evaluations.
