Dual selections based knowledge transfer learning for cross-subject motor imagery EEG classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery electroencephalography (MI-EEG) is crucial for non-invasive brain-computer interfaces (BCIs).
• Cross-subject MI-EEG classification faces challenges due to sample variability, limiting BCI applications.
• Existing domain adaptation methods struggle with redundant features and inaccurate pseudo-labels.

Purpose of the Study:

• To propose a novel method, Dual Selections based Knowledge Transfer Learning (DS-KTL), for high-performance cross-subject MI-EEG classification.
• To address limitations of existing methods in feature selection and pseudo-label correction within the target domain.

Main Methods:

• DS-KTL employs centroid alignment and Riemannian tangent space features for adaptation.
• The method incorporates dual selections with regularizations during feature adaptation to enhance iterative classification.
• Feature adaptation and pseudo-label correction are key components of the DS-KTL approach.

Main Results:

• DS-KTL demonstrates feasibility and effectiveness on benchmark MI-EEG datasets for cross-subject classification.
• The method achieves significant performance improvements compared to state-of-the-art techniques.
• Ablation studies validate the effectiveness of DS-KTL's characteristics and parameters.

Conclusions:

• DS-KTL offers a robust solution for cross-subject MI-EEG classification in BCI.
• The proposed method effectively handles feature redundancy and pseudo-label inaccuracies.
• DS-KTL shows promise for advancing non-invasive BCI development.