Convolutional Neural Network for Drowsiness Detection Using EEG Signals.

Area of Science:

• Biomedical signal processing
• Machine learning applications in healthcare
• Neuroscience and brain-computer interfaces

Background:

• Drowsiness detection (DD) is critical for preventing accidents, especially in safety-sensitive occupations.
• Electroencephalogram (EEG) signals offer a reliable physiological measure for assessing brain activity related to fatigue.
• Existing deep learning (DL) approaches for EEG-based DD require further optimization for accuracy and generalization.

Purpose of the Study:

• To propose and evaluate a novel EEG classification system for drowsy detection using deep learning.
• To investigate the efficacy of a convolutional neural network (CNN) architecture for fatigue monitoring.
• To enhance the DD system's performance through data augmentation and comparative analysis of DL frameworks.

Main Methods:

• Utilized a wearable Emotiv EPOC+ headset for collecting 14-channel EEG data.
• Implemented a data augmentation strategy to mitigate overfitting and improve model robustness.
• Developed and tested a CNN model using the Keras library for EEG signal classification.
• Conducted a comparative study to validate the chosen DL architecture and framework.

Main Results:

• The proposed EEG-based DD system achieved a high accuracy of 90.42% in discriminating between drowsy and awake states.
• Demonstrated the effectiveness of the data augmentation technique in enhancing classification performance.
• The CNN architecture proved efficient for analyzing EEG signals for fatigue detection.

Conclusions:

• The developed DL-based EEG classification system offers a promising solution for accurate drowsy detection.
• The study highlights the potential of CNNs and data augmentation in improving fatigue monitoring systems.
• The proposed methodology provides a robust framework for future research in real-time drowsiness assessment.