Efficient EEG-Based Person Identification: A Unified Framework from Automatic Electrode Selection to Intent...

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Electroencephalography (EEG) is crucial for human-computer interaction, with person identification being a key application.
• Existing EEG-based identification methods face challenges in pipeline design, automatic electrode selection, feature extraction, and intent recognition.

Purpose of the Study:

• To propose a unified deep learning framework integrating automatic electrode selection, person identification, and intent recognition.
• To develop a novel backbone network (AES-MBE) for enhanced feature extraction and intent recognition.

Main Methods:

• A novel deep learning framework incorporating automatic electrode selection (AES) and a multi-scale bidirectional encoder (MBE) with channel attention.
• Validation on the PhysioNet EEG Motor Movement/Imagery Dataset (EEGMMIDB) with 109 subjects across 4 tasks.

Main Results:

• Achieved 98.82% accuracy in person identification using only 4 electrodes.
• Attained an average intent recognition accuracy of 91.58%.
• Demonstrated strong stability and robustness across varying numbers of users.

Conclusions:

• The proposed unified framework effectively integrates EEG-based person identification and intent recognition.
• The AES-MBE network enhances feature extraction by adaptively capturing local and global temporal dependencies.
• The approach offers a robust and efficient solution for advanced EEG-based applications.