基于混合脑机接口多模态特征图融合的认知工作负荷识别OA

The recognition task of cognitive workload holds significant importance in the field of brain-computer interfaces.Existing methods rely on single-modal electroencephalogram data and use shallow graph neural networks,which limits the performance of cognitive workload recognition.Traditional feature fusion methods struggle to effec-tively combine the high spatial resolution of near-infrared spectroscopy with the high temporal resolution of EEG sig-nals.This study proposes a novel architecture based on LGCN-DATF(Local Graph Convolutional Dual-Channel A-daptive Transformer Feature Map Fusion)aiming to simulate the complexity of dynamic changes between different brain regions.It captures the real-time connection status between signal channels by introducing a dynamic adjacency matrix.A unique brain graph learning module is designed,which is conducive to a deeper understanding and predic-tion of the dynamic changes in cognitive workload.The proposed model is tested under two training strategies.The re-sults show that the accuracy rates of subject-dependent training in mental arithmetic and working memory tasks are 87.3%and 89.1%,respectively,while the accuracy rates of subject-independent training are 68.5%and 55.6%,respec-tively.These results verify that the proposed model can effectively recognize cognitive workload in complex environments.