Brain power mapping with optimized deep learning for EEG-based pilot fatigue detection

With the surge in flight missions, pilot fatigue has emerged as a significant aviation safety concern. Accurate and timely detection of pilot fatigue is essential for preventing flight accidents. Electroencephalography (EEG) is widely recognized as a “gold standard” for detecting fatigue. However, current EEG-based methods face challenges regarding accuracy and efficiency. This study introduces a novel approach combining an EEG-based brain power map (BPM) with an improved sparrow search algorithm (ISSA) optimized deep learning model for pilot fatigue detection. Specifically, for fatigue feature generation, we project the electrode positions from 3D space to 2D space using equidistant azimuthal projection, and the indicators of (θ + α)/β, (θ + α)/(α + β), and θ/β are selected as features. The bicubic interpolation algorithm is used to generate BPM. For the detection model design, we propose a hybrid model that integrates convolutional neural network (CNN), Bidirectional Mogrifier-gated recurrent unit (BiMGRU), and Multi-head attention mechanism (MHAM) to extract spatial and correlated features from BPM for fatigue recognition. What's more, this paper introduces an ISSA with a sine–cosine strategy and a Gaussian-Cauchy mutation strategy to optimize critical hyperparameters synchronously in model training, which reduces the reliance on empirical parameters selection, enhancing both the efficiency and accuracy of network training. Experiments on EEG data collected from 18 pilots during simulated flight tasks, where each pilot provides data across four states, and the results show that the proposed approach achieves the accuracy and F1_score both of 98.75 % for fatigue detection, outperforming state-of-the-art approaches, highlighting its significant practical application value.