Research on MI EEG signal classification algorithm using multi-model fusion strategy coupling.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Accurate recognition of motor imagery (MI) electroencephalography (EEG) signals is crucial for brain-computer interface (BCI) development.
• Existing feature extraction and classification methods may not fully capture the complexity of MI EEG signals.

Purpose of the Study:

• To enhance the accuracy of MI EEG signal recognition.
• To explore advanced feature extraction, selection, and classification techniques for improved BCI performance.

Main Methods:

• Employed power spectral density and wavelet packet decomposition combined with common spatial pattern for in-depth feature extraction.
• Utilized F-test for feature selection and Platt Scaling for probability calibration of six basic classifiers (RF, SVM, LR, GNB, XGBoost, LightGBM).
• Performed model fusion on the top-performing classifiers.

Main Results:

• The proposed method achieved an average accuracy of 91.46% on Datasets 2a of the 4th International BCI Competition across nine subjects.
• Feature fusion and classifier calibration significantly improved the classification accuracy of MI EEG signals.

Conclusions:

• Model fusion is an effective strategy for enhancing MI EEG signal classification accuracy.
• The study provides valuable insights and a reference for future research in MI-based BCI systems.