Band power feature part-based convolutional neural network with African vulture optimization fostered channel...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalogram (EEG) based motor imagery (MI-EEG) classification is crucial for brain-computer interfaces (BCIs).
• Acquiring EEG signals with numerous channels poses challenges for real-world BCI applications.
• Selecting optimal EEG channel subsets without compromising classification performance is a significant problem.

Purpose of the Study:

• To propose an efficient channel selection method for EEG classification in BCIs.
• To enhance the accuracy and reduce computational cost of MI-EEG classification.
• To improve the feasibility of real-time BCI applications.

Main Methods:

• A novel PCNNC-AVOACS-EEG method combining a band power feature part-based convolutional neural network (PCNNC) with African vulture optimization (AVO) for channel selection was developed.
• EEG signals from BCI Competition IV, dataset 1 were pre-processed using contrast-limited adaptive histogram equalization and feature extracted via hexadecimal local adaptive binary pattern (HLABP).
• HLABP extracted alpha and beta band features, with band power data serving as input for the PCNNC, while AVO optimized channel selection.

Main Results:

• The proposed PCNNC-AVOACS-EEG technique achieved higher classification accuracy and area under the curve compared to existing methods.
• Significant reductions in computation time were observed, with improvements of 70%, 60%, and 65.714% in different experimental settings.
• The method demonstrated enhanced categorization accuracy on the test set, a key metric for real-time BCI.

Conclusions:

• The PCNNC-AVOACS-EEG method effectively addresses the challenge of channel selection in MI-EEG classification.
• This approach offers a promising solution for developing more practical and efficient real-time BCI systems.
• The optimization technique significantly improves classification performance and reduces computational load.