Motor imagery electroencephalogram classification algorithm based on joint features in the spatial and frequency...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery electroencephalography (MI-EEG) is crucial for brain-computer interfaces (BCIs) and rehabilitation.
• Challenges include small single-subject training data and significant inter-subject variability, leading to low classification accuracy and poor generalization.
• Existing MI classification models struggle with these limitations.

Purpose of the Study:

• To propose a novel electroencephalography (EEG) joint feature classification algorithm for MI tasks.
• To address the limitations of small sample sizes and individual differences in MI-EEG data.
• To improve the accuracy and generalization ability of MI classification.

Main Methods:

• A joint feature classification algorithm using instance transfer and ensemble learning is proposed.
• EEG data undergoes preprocessing, followed by feature extraction using Common Spatial Patterns (CSP) and Power Spectral Density (PSD).
• An ensemble learning model combining Kernel Mean Matching (KMM) and Transfer Learning Adaptive Boosting (TrAdaBoost) is employed for classification.

Main Results:

• The algorithm achieved an average accuracy of 91.5% on the BCI Competition IV Dataset 2a.
• The algorithm achieved an average accuracy of 83.7% on the BCI Competition IV Dataset 2b.
• Performance significantly surpassed other compared algorithms, demonstrating effectiveness and stability.

Conclusions:

• The developed algorithm effectively utilizes EEG signals and enriches features for improved MI signal recognition.
• It offers a new, robust approach to overcoming challenges in MI classification.
• The method shows significant potential for advancing BCI and rehabilitation technologies.