军事知识服务平台
3D position embedding and hierarchical cross-graph fusion for EEG emotion recognition
摘要 Electroencephalography(EEG)signals exhibit spatial sparsity and nonlinearity,making the precise modeling of inter-channel relationships crucial for emotion recognition.Although graph-based approaches have shown great potential,challenges persist:how to effectively integrate the stable spatial layout of electrodes with the dynamically changing functional connectivity during emotional tasks,thereby constructing models that combine discriminative power with neurophysiological inter-pretability.To address this,we propose a three-dimension(3D)position-embedding graph attention long-term memory model(PEGLM),which introduces a hierarchical cross-transformer fusion mod-ule to adaptively integrate the 3D positional graph of electrodes with a functional connectivity graph based on phase-locking value(PLV).This captures the synergistic characteristics of brain anatomi-cal structure and functional interaction.We further employ a graph attention network(GAT)to ex-tract spatial-frequency features using differential entropy(DE)and power spectral density(PSD),and a bidirectional long short-term memory(Bi-LSTM)with attention to model their temporal evolu-tion.Experiments on DEAP,SEED,and SEED-IV datasets achieve accuracies of 98.17%,97.37%,and 94.04%,respectively,outperforming existing graph-based methods.The results demonstrate that PEGLM not only achieves superior performance,but also provides neurophysiologi-cal insights into the spatial-functional coupling mechanisms during emotional processing.
年,卷(期): 2026, 32(3)
机标分类号: TP391.1 TN911 R395.1
在线出版日期: 2026年9月28日