tart All work

RESEARCH / 2022 PREPRINT

EEG-BBNet

Can brain signals identify a person?

Co-author · August 2022

My contribution

Ratha Rotruchiphong is listed on the preprint with Kamnoetvidya Science Academy affiliation. This was collaborative research; individual implementation responsibilities are not specified in the paper.

Download CV (English)

Try it · mini version

See what it can do

Paper architecture: EEG feature extraction and connectivity graphs feed graph convolution layers and a subject classifier.
Original architecture figure · Lakhan et al., EEG-BBNet, arXiv v1 (2022).

KVIS research · Co-authored arXiv preprint · August 2022

What this project does

EEG-BBNet investigates identifying people from brain signals. The research combines convolutional features from EEG recordings with graph connectivity between electrodes to classify subject identity.

Inside the project

  • Study a hybrid CNN and graph convolutional framework for EEG biometrics.
  • Compare electrode-connectivity measures and experimental task settings.
  • Evaluate recognition under same-session and across-session conditions.

Skills & practical experience

Research exposure to biosignal processing, graph-based machine learning, experimental comparisons, and communicating results with appropriate limitations.

EEG signal processingCNNsGraph neural networksBrain biometricsExperimental evaluationResearch collaboration

I am a named co-author on the 2022 preprint. The paper does not allocate individual implementation tasks, so these are research areas covered by the collaboration.

Project context

The framework combines convolutional EEG features with graph neural networks representing relationships between electrodes.

54OpenBMI participants
62EEG electrodes
3Experimental tasks

Signals → connections → identity

Motor imagery, event-related potentials (ERP), and visual stimulation provide three evaluation settings. Connectivity measures describe how EEG channels relate.

Original paper figure comparing electrode placements and five connectivity matrices
Original electrode and connectivity figure · Lakhan et al., 2022. Tap to enlarge.
SELECTED RESULT / TABLE III99.26% ± 0.41

ERP, session I, Pearson correlation connectivity. Correct recognition rate: mean ± standard deviation across five folds.

Across-day performance drops without fine-tuning; these controlled results do not establish real-world biometric reliability.

This showcase covers the 2022 preprint. The later IEEE Sensors Letters article has a revised author list that does not include Ratha.

Figures reproduced without modification from Lakhan et al., arXiv:2208.08901v1 · CC BY-NC-ND 4.0.