CNN based classification of motor imaginary using variational mode decomposed EEG-spectrum image.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Motor imagery (MI) recognition is crucial for brain-computer interfaces.
• Effective preprocessing of electroencephalography (EEG) signals is essential for accurate MI classification.
• Current methods may not fully capture the complex temporal and frequency information within EEG data.

Purpose of the Study:

• To propose a novel approach for EEG signal preprocessing using spectrum image generation.
• To evaluate the effectiveness of this approach for Convolutional Neural Network (CNN) based Motor Imagery (MI) recognition.
• To compare the performance of different CNN architectures using the generated EEG spectrum images.

Main Methods:

• Extracting Variational Mode Decomposition (VMD) modes from EEG signals.
• Applying Short Time Fourier Transform (STFT) to VMD modes to create EEG spectrum images.
• Utilizing CNN architectures (EEGNet, DeepConvNet, AlexNet, LeNet) for classification of MI tasks.

Main Results:

• EEG spectrum images generated using VMD-STFT were used as input for CNNs.
• EEGNet achieved high average accuracies (e.g., 91.37%, 94.41%) across four datasets.
• The proposed method demonstrated superior performance compared to existing literature.

Conclusions:

• The VMD-STFT method for generating EEG spectrum images is a promising technique for time-frequency analysis.
• This approach enhances the effectiveness of CNN-based MI recognition.
• The findings suggest a significant advancement in EEG signal processing for brain-computer interfaces.