文献类型：期刊文献

英文题名：AGCTN: adaptive graph convolutional transformer network for hyperscanning EEG mental workload recognition

作者：Chen, Lan-lan[1];Xu, Jia-min[1];Zhou, Shu-jin[2];Zhang, Ming-ming[3]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, 130 Meilong Rd, Shanghai 200237, Peoples R China;[2]Shanghai Normal Univ, Shanghai Inst Early Childhood Educ, Shanghai 200234, Peoples R China;[3]Shanghai Normal Univ, Sch Psychol, Shanghai 200234, Peoples R China

年份：2026

卷号：330

外文期刊名：EXPERT SYSTEMS WITH APPLICATIONS

收录：;EI(收录号：20262320865031);WOS:【SCI-EXPANDED(收录号:WOS:001796487500001)】；

基金：This work was supported in part by the National Natural Science Foundation of China (No. 62376095) .

语种：英文

外文关键词：Hyperscanning; Synchronization; Graph convolutional networks; Transformer; Mental workload recognition

摘要：Traditional mental workload methods have primarily focused on individual brain analysis, overlooking the interactions among multiple brains. With the advancement of hyperscanning technology, increasing attention has been paid to the understanding of inter-brain interactions and their influence on cognitive load during collaborative tasks. This paper introduces an innovative framework for EEG mental workload recognition, i.e., the Adaptive Graph Convolutional Transformer Network (AGCTN), aimed at addressing the challenges of mental workload assessment in multi-brain collaborative tasks. By integrating Graph Convolutional Networks (GCNs) and Transformers, AGCTN effectively captures local spatial features and global dependencies in EEG signals. Specifically, an adaptive adjacency matrix that dynamically adjusts the spatial relationships between EEG signals enriches the expressive power of GCN, while a Transformer module with cross-attention mechanism can effectively model long-range dependencies across brain signals. Besides, a Gated Feature Fusion (GFF) module is introduced to prioritize more discriminative features while suppressing redundancy. This framework overcomes the shortcomings of conventional approaches in handling complex spatial dependencies in multi-brain scenarios. Evaluations on three multi-brain EEG datasets demonstrated the superiority of AGCTN over state-of-the-art models in mental workload recognition. Further analyses revealed that inter-brain networks outperformed their intra-brain counterparts in cognitive load recognition, suggesting that social interactions prompted more discriminative neural patterns through inter-brain synchronization. This scheme can be developed into an online warning module for safety-critical human-machine systems, enabling real-time assessment of operators' cognitive states through dynamic decoding of inter-brain coupling. Moreover, it can support adaptive systems to autonomously regulate task demands based on team workload.