TCANet: a temporal convolutional attention network for EEG-based cognitive load assessment in simulated flight tasks

Effective cognitive load assessment facilitates optimized task allocation and plays a critical role in evaluating human performance in domains such as flight safety monitoring and human-computer interaction. Electroencephalography (EEG) has been extensively employed in cognitive load research owing to its high temporal resolution, non-invasiveness, and accessibility. However, EEG signals exhibit substantial variability across individuals and task conditions, making it challenging to extract stable and discriminative features related to cognitive load. Accordingly, we propose TCANet, a Temporal Convolutional Attention Network designed to learn discriminative EEG representations for cognitive load classification in simulated flight tasks. The proposed network integrates multi-branch temporal convolutions, dilated convolutions, and attention mechanisms to model temporal, spatial, and channel-wise EEG information while reducing redundant feature responses. We designed an EEG-based cognitive load classification paradigm involving two pilot-simulated flight scenarios. A total of 20 participants were recruited, with 15 participants completing the turning task and 5 participants completing the landing task. TCANet achieved average classification accuracies of 90.15% and 81.52% on the turning-task and landing-task datasets, respectively. The turning-task dataset served as the primary evaluation dataset, whereas the landing-task dataset provided a supplementary evaluation under a different simulated flight task. Statistical analyses of frequency-band power indicated task-condition-related variations in EEG spectral activity. Neurophysiological variations, particularly in the Theta and Alpha bands, and feature representation analyses further provided interpretable evidence regarding cognitive-load-related EEG patterns and the discriminative representations learned by TCANet.