Capped L21-norm-based common spatial patterns for EEG signals classification applicable to BCI systems.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Common Spatial Patterns (CSP) is effective for electroencephalogram (EEG) signal classification.
• Conventional CSP's reliance on the L2-norm makes it susceptible to outliers and noise.
• Robustness is crucial for reliable EEG signal analysis.

Purpose of the Study:

• To develop a more robust CSP algorithm for EEG signal classification.
• To mitigate the impact of outliers and noise on CSP performance.
• To introduce the capped L21-norm-based common spatial patterns (CCSP-L21) method.

Main Methods:

• Proposed a novel CCSP-L21 method utilizing the capped L21-norm instead of the L2-norm.
• The capped L21-norm enhances robustness by incorporating L1-norm properties and mitigating extreme outliers.
• Developed a non-greedy iterative procedure to optimize the proposed objective function.

Main Results:

• The CCSP-L21 method demonstrated superior performance on three real-world BCI competition datasets.
• Achieved the highest average recognition rates: 91.67%, 85.07%, and 82.04% on the respective datasets.
• The proposed method effectively reduces the influence of noise and outliers in EEG signal classification.

Conclusions:

• CCSP-L21 offers a robust and effective approach for EEG signal classification.
• The capped L21-norm provides significant advantages over the traditional L2-norm for noisy EEG data.
• This robust model holds promise for advancing Brain-Computer Interface (BCI) applications.