Selective Feature Generation Method Based on Time Domain Parameters and Correlation Coefficients for Filter-Bank-CSP...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Motor imagery (MI) classification is crucial for brain-computer interfaces (BCIs).
• Existing channel selection methods for MI classification may not be optimal.
• Filter-bank common spatial pattern (FBCSP) is a powerful feature extraction technique for MI.

Purpose of the Study:

• To propose a novel MI classification algorithm with an improved channel selection strategy.
• To enhance the classification performance of BCIs by optimizing feature extraction.

Main Methods:

• A new channel selection algorithm based on the Fisher ratio of time domain parameters (TDPs) and correlation coefficients.
• Identification of a 'principle channel' and a 'supporting channel set' for FBCSP feature generation.
• Evaluation using BCI Competition III Dataset IVa and BCI Competition IV Dataset I.

Main Results:

• The proposed algorithm significantly improved MI classification performance.
• Utilizing FBCSP features from the selected supporting channel set enhanced accuracy.
• Demonstrated effectiveness on datasets with varying numbers of channels (18 and 59).

Conclusions:

• The proposed MI-relevant channel selection method is effective for enhancing FBCSP-based classification.
• This approach offers a promising advancement for BCI system performance.
• The method is robust and validated on standard BCI datasets.