SET Net A Hybrid Deep Learning Framework For EEG Based Attention Relaxation Classification In Brain Computer...

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain Computer Interface (BCI) systems enable direct brain-environment communication for assistive technologies.
• Electroencephalography (EEG) is a standard non-invasive method for neural activity acquisition in BCIs.
• Classifying cognitive states like attention and relaxation from noisy, non-stationary EEG signals remains challenging.

Purpose of the Study:

• To propose a novel hybrid deep learning architecture, the Squeeze-Excitation (SE) Transformer Network (SET-Net).
• To enhance the classification accuracy of cognitive states, specifically attention and relaxation, using EEG signals.
• To develop a scalable system for real-time BCI applications.

Main Methods:

• EEG signals were preprocessed and segmented into temporal windows.
• Spectrogram representations of the EEG signals were analyzed.
• A hybrid deep learning architecture, SET-Net, incorporating Squeeze-Excitation modules and transformer networks, was developed and applied.

Main Results:

• The proposed SET-Net achieved a classification accuracy of 93.7%.
• The model demonstrated a high F1-score of 0.93 and ROC-AUC of 0.98.
• The results indicate enhanced discrimination of EEG-based attention-relaxation states.

Conclusions:

• The hybrid SET-Net model effectively overcomes the limitations of noisy and non-stationary EEG signals.
• The developed system shows significant potential for real-time BCI applications, particularly in assistive technology and human-computer interaction.
• This research contributes to advancing the capabilities of EEG-based BCI systems.