A Novel Hybrid Approach for Drowsiness Detection Using EEG Scalograms to Overcome Inter-Subject Variability

Area of Science:

• Neuroscience
• Machine Learning
• Occupational Safety

Background:

• Drowsiness is a major risk factor for accidents in various industries.
• Electroencephalography (EEG) offers direct brain activity measurement for drowsiness detection.
• EEG signal non-stationarity and inter-subject variability pose challenges for accurate detection.

Purpose of the Study:

• To develop a robust drowsiness detection method using EEG signals.
• To address challenges of EEG signal variability and improve detection accuracy.
• To compare a novel hybrid CNN-SVM approach with existing methods.

Main Methods:

• A hybrid framework combining Convolutional Neural Networks (CNNs) for feature extraction and Support Vector Machines (SVMs) for classification.
• Utilized Continuous Wavelet Transform (CWT) to generate 2D EEG scalograms for CNN feature extraction.
• Compared the proposed CNN-SVM model with 1D CNNs and transfer learning models (VGG16, ResNet50) on the DROZY dataset.

Main Results:

• The hybrid CNN-SVM model achieved a high accuracy of 98.33% in drowsiness detection.
• The proposed method significantly outperformed 1D CNNs and transfer learning models.
• The approach demonstrated effectiveness in minimizing inter-subject variability.

Conclusions:

• The hybrid CNN-SVM approach offers a robust and accurate solution for EEG-based drowsiness detection.
• This method has significant potential for enhancing safety in high-risk occupational settings.
• The use of 2D EEG scalograms with CNNs is a promising direction for future research.