A novel method for EEG-based motor imagery classification using feature fusion.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery (MI) classification from electroencephalography (EEG) signals is crucial for brain-computer interfaces.
• Existing methods often struggle to fully capture the complex spectral-temporal-spatial characteristics of EEG data.

Purpose of the Study:

• To introduce a novel multi-scale feature fusion framework for enhanced EEG-based MI classification.
• To leverage the nonlinear intrinsic characteristics and convolutional features of EEG data.

Main Methods:

• Developed a multi-scale feature fusion framework integrating spectral-temporal-spatial information.
• Employed factor analysis (FA) for dimensionality reduction and common spatial pattern (CSP) for channel selection.
• Utilized a support vector machine (SVM) classifier.

Main Results:

• Proposed feature fusion models outperformed current state-of-the-art MI classification systems.
• An SVM model achieved 86.92% accuracy on the BCIC-IV-2a dataset.
• Selecting 12 channels using CSP resulted in a superior accuracy of 88.17% compared to using all 22 or 8 channels.

Conclusions:

• The multi-scale feature fusion framework effectively captures complex EEG characteristics for improved MI classification.
• Optimized channel selection significantly enhances classification performance.
• This approach offers a promising direction for developing more accurate brain-computer interfaces.