Method for EEG signal recognition based on multi-domain feature fusion and optimization of multi-kernel extreme...

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Existing multi-class motor imagery recognition methods suffer from limited feature extraction and low classification accuracy.
• Effective feature extraction and robust classification are crucial for advancing brain-computer interface (BCI) systems.

Purpose of the Study:

• To develop an improved Electroencephalogram (EEG) signal recognition method for multi-class motor imagery.
• To enhance classification accuracy and efficiency in BCI applications through advanced feature engineering and machine learning.

Main Methods:

• EEG signal preprocessing using Improved Comprehensive Ensemble Empirical Mode Decomposition (ICEEMD) and Pearson correlation coefficient.
• Multi-domain feature extraction (time, frequency, spatial) using multivariate autoregressive (MVAR) models, wavelet packet decomposition, and Riemannian geometry.
• Feature fusion and dimensionality reduction via Kernel Principal Component Analysis (KPCA).
• Classification using a Radius-incorporated multi-kernel extreme learning machine (RIO-MKELM).

Main Results:

• The proposed method effectively fuses multi-domain features, improving feature selection and retaining crucial information.
• Achieved high classification accuracy of 95.49%, with sensitivity at 97.88%, specificity at 98.12%, recall at 97.88%, and F1 Score at 96.67%.
• Demonstrated superior performance compared to existing methods in motor imagery recognition.

Conclusions:

• Multi-domain feature fusion combined with an optimized RIO-MKELM significantly enhances EEG signal recognition.
• The developed method offers a promising approach for practical and effective brain-computer interface (BCI) system development.
• This research contributes to overcoming limitations in current BCI technologies, paving the way for more sophisticated applications.