Automated recognition of mental cognitive workload through electroencephalography and a deep learning approach

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Traditional cognitive workload identification methods have limitations.
• Electroencephalography (EEG) signals offer a non-invasive measure of brain activity.
• Accurate workload assessment is crucial for performance and safety.

Purpose of the Study:

• To develop an automated system for cognitive workload identification using EEG signals.
• To overcome limitations of existing workload identification techniques.
• To integrate deep learning with advanced signal processing for improved accuracy.

Main Methods:

• A pre-trained Convolutional Neural Network (CNN), the Xception network, was utilized.
• Functional connectivity matrices were generated using the corrected imaginary phase locking value (ciPLV) method.
• The CNN processed the ciPLV matrices, and its output was combined with nonlinear dynamic features.
• A feedforward neural network performed the final classification.

Main Results:

• The integrated system achieved high accuracy in workload identification.
• Over 98% accuracy was obtained for distinguishing between two workload levels.
• 92.50% accuracy was achieved for differentiating three workload levels.
• The approach demonstrated the effectiveness of combining deep learning with advanced EEG features.

Conclusions:

• The proposed integrated deep learning system shows significant promise for automatic cognitive workload identification.
• This novel approach offers a more accurate and efficient method compared to traditional techniques.
• The findings highlight the potential of advanced signal processing and machine learning in understanding cognitive states.