Real -Time EEG-Based Detection of Cognitive Fatigue in Human-Machine Interaction Systems: A Biomedical Engineering Approach

International audience Cognitive fatigue significantly impairs human performance in safety-critical Human-Machine Interaction (HMI) environments such as driving, aviation, and industrial control. Traditional detection methods based on behavioral or selfreport measures are subjective and non-continuous. This study presents a real-time Electroencephalography (EEG)-based cognitive fatigue detection system integrating biomedical signal processing and deep learning for accurate and rapid assessment of mental fatigue. EEG data were acquired from a 14-channel headset and preprocessed using bandpass and notch filters, followed by Independent Component Analysis for artifact removal. Time-frequency features such as Power Spectral Density (PSD), Hjorth parameters, and the (Theta + Alpha)/Beta ratio were extracted and classified using a hybrid Convolutional-Long Short-Term Memory (CNN-LSTM) model. The system achieved a classification accuracy of 94.2%, outperforming traditional models such as SVM and Random Forest. Real-time implementation on an embedded platform achieved a processing latency of <500 ms, confirming operational feasibility. The results demonstrate the potential of biomedical EEG monitoring for continuous cognitive state assessment and adaptive automation in high-demand environments.