Multi-source domain adaptation based tempo-spatial convolution network for cross-subject EEG classification in RSVP...

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Subject-dependent electroencephalogram (EEG) classification methods for rapid serial visual presentation (RSVP) tasks require extensive subject-specific data and lengthy calibration.
• Cross-subject EEG classification offers reduced calibration but remains a significant challenge in RSVP tasks.

Purpose of the Study:

• To develop an effective cross-subject EEG classification method for RSVP tasks that minimizes calibration requirements.
• To enhance the generalization and accuracy of EEG classification across different subjects.

Main Methods:

• Proposed a multi-source domain adaptation based tempo-spatial convolution (MDA-TSC) network.
• Employed multi-scale tempo-spatial convolution for extracting domain-invariant features.
• Utilized multi-branch domain-specific feature extraction and alignment to address inter-subject variability.
• Incorporated domain-specific classifiers for optimized prediction on target domains.

Main Results:

• The MDA-TSC network demonstrated superior performance compared to existing methods on a benchmark RSVP dataset for cross-subject classification.
• Ablation studies and visualization confirmed the network's effectiveness and robustness.
• Achieved significant improvements in cross-subject classification accuracy.

Conclusions:

• The proposed MDA-TSC network effectively enhances cross-subject EEG classification in RSVP tasks.
• This approach has the potential to substantially reduce system calibration time for practical applications.