Proceedings of the Annual Meeting of the Cognitive Science Society

Mental fatigue among drivers is a primary factor in many traffic accidents. Electroencephalography (EEG), which directly measures neurophysiological activities in the brain, is commonly used for fatigue recognition. However, cross-subject research in fatigue recognition using EEG faces challenges such as low spatial resolution and significant individual variability. Inspired by neuroscience, a dynamic graph convolution learning from functional neuroimaging (FNI-DGCNN) is proposed, making up for EEG's low spatial resolution. We first use a multi-scale spatiotemporal learning block to extract EEG features with attention allocation, then initialize the adjacency matrix based on prior knowledge about fatigue recognition mechanisms from functional neuroimaging, use the extracted features and the adjacency matrix to initialize the graph, and finally use dynamic graph convolution further to study the intrinsic functional connectivity of mental fatigue. The proposed method achieves an accuracy of 88.89% among 17 subjects, outperforming existing EEG models for cross-subject.