A single-joint multi-task motor imagery EEG signal recognition method based on Empirical Wavelet and Multi-Kernel...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Classifying electroencephalography (EEG) signals for multi-task Brain-Computer Interfaces (BCIs) is crucial for finer command control.
• Distinguishing between EEG signals from the same joint during motor imagery tasks presents a significant challenge due to overlapping spatial distributions.

Purpose of the Study:

• To propose and validate a novel method for recognizing single-joint, multi-task motor imagery EEG signals.
• To enhance the differentiation of EEG signals from similar motor tasks for improved BCI performance.

Main Methods:

• EEG signals from three motor imagery tasks (wrist extension, flexion, abduction) were collected from six participants.
• Empirical Wavelet Decomposition (EWT) was employed for signal decomposition, screening, and reconstruction.
• Common Spatial Patterns (CSP) were used for feature extraction, followed by classification with a Multi-Kernel Extreme Learning Machine (MKELM).

Main Results:

• EWT processing significantly enhanced the time and frequency differences between EEG signals of different motor imagery classes.
• The MKELM model achieved an average recognition accuracy of 91.93% for the single-joint multi-task EEG signals.

Conclusions:

• The proposed EWT-MKELM method demonstrates superior performance in differentiating complex EEG signals compared to other decomposition techniques like EMD, VMD, LMD, and WPD.
• MKELM outperformed traditional machine learning and deep learning models in recognition accuracy and training speed.
• This research offers a promising new approach for developing more precise and responsive Brain-Computer Interface commands.