Classification of Motor Imagery Electroencephalography Signals Based on Image Processing Method

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interface (BCI) technology, particularly motor imagery electroencephalography (MI-EEG), is rapidly advancing.
• Classifying MI-EEG signals accurately remains a significant challenge due to signal non-stationarity and complex characteristics.
• Existing frameworks often struggle with noise interference and extracting relevant temporal and frequency features.

Purpose of the Study:

• To propose a novel deep learning framework, IS-CBAM-CNN, to improve the accuracy of MI-EEG classification.
• To address the non-stationary nature, temporal localization, and frequency band distribution of MI-EEG signals.
• To enhance feature extraction and reduce noise interference for more robust MI-EEG pattern recognition.

Main Methods:

• Utilized image subtraction (IS) on C3 and C4 channels of MI-EEG signals to reduce redundancy and increase feature differences.
• Integrated a Convolutional Block Attention Module (CBAM) into a convolutional neural network (CNN) base classifier.
• Employed CBAM to adaptively extract temporal and frequency distribution information, minimizing noise and enhancing pattern robustness.

Main Results:

• The proposed IS-CBAM-CNN framework achieved a mean accuracy of 79.6% on the BCI competition IV dataset 2b.
• The average kappa value reached 0.592, indicating a significant improvement in classification performance.
• Experimental results demonstrated the feasibility and effectiveness of the proposed framework in enhancing MI-EEG signal classification.

Conclusions:

• The IS-CBAM-CNN framework effectively addresses key challenges in MI-EEG signal processing.
• The integration of image subtraction and CBAM significantly improves classification accuracy and robustness.
• This study validates a promising approach for advancing MI-EEG-based brain-computer interface technology.