Channel-Dependent Multilayer EEG Time-Frequency Representations Combined with Transfer Learning-Based Deep CNN...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalogram (EEG) signal analysis is vital for brain function studies, clinical diagnostics, and brain-computer interfaces.
• Recognizing motor imagery EEG signals with limited channels is crucial for developing portable and real-time applications.
• Data sparsity and feature extraction challenges exist in low-channel count EEG scenarios.

Purpose of the Study:

• To propose a novel framework for enhanced motor imagery EEG signal classification using few channels.
• To develop a method that effectively represents multidimensional information from limited EEG channels.
• To leverage deep learning and transfer learning for improved classification accuracy in sparse, few-channel EEG data.

Main Methods:

• A continuous wavelet transform converts time-domain EEG signals into 2D time-frequency representations.
• Channel-dependent multilayer EEG time-frequency representations (CDML-EEG-TFR) are created by concatenating these images.
• A deep convolutional neural network with EfficientNet backbone, utilizing pre-trained weights for transfer learning, is employed to analyze CDML-EEG-TFR.

Main Results:

• The proposed framework successfully integrates temporal, spatial, and channel features from CDML-EEG-TFR.
• Transfer learning effectively mitigates data sparsity issues inherent in few-channel EEG data.
• Experimental results on the BCI Competition IV 2b dataset achieved a classification accuracy of 80.21% for motor imagery EEG signals.

Conclusions:

• The CDML-EEG-TFR combined with EfficientNet-based transfer learning offers a powerful approach for few-channel EEG signal classification.
• This method significantly enhances classification accuracy, demonstrating its potential for practical applications.
• The study provides a foundation for future research in medical and sports-related fields utilizing low-channel EEG systems.