引用本文:王雨佳,鞠翔宇,于扬,李明.基于脑电频谱时空特征的认知负荷评估[J].控制理论与应用,2025,42(1):50~58.[点击复制]
WANG Yu-jia,JU Xiang-yu,YU Yang,LI Ming.Cognitive workload assessment based on temporal and spatial characteristics of electroencephalogram spectrum[J].Control Theory and Technology,2025,42(1):50~58.[点击复制]
基于脑电频谱时空特征的认知负荷评估
Cognitive workload assessment based on temporal and spatial characteristics of electroencephalogram spectrum
摘要点击 2201  全文点击 23  投稿时间:2023-04-19  修订日期:2024-11-22
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2023.30238
  2025,42(1):50-58
中文关键词  认知负荷评估  脑电频谱图  时空特征  时空双路神经网络
英文关键词  cognitive workload assessment  power spectrum pattern  temporal and spatial characteristics  dual-stream network for spatial and temporal representation
基金项目  国家自然科学基金项目(62076248)资助.
作者单位邮编
王雨佳 国防科技大学智能科学学院 410073
鞠翔宇 国防科技大学智能科学学院 
于扬 国防科技大学智能科学学院 
李明* 国防科技大学智能科学学院 410073
中文摘要
      准确的认知负荷评估对于增强人机协调能力、提升人机共融系统效率具有重要意义. 目前常用的基于脑电信号的认知负荷评估技术中, 在提取和利用频谱信息的时间和空间关系等方面普遍缺少有效手段. 本文针对脑电频谱图空间分辨率低的特点, 引入胶囊网络有效表征不同导联之间的相对空间关系; 针对脑电频谱图随时间波动的特点, 设计了由长短期记忆网络等组成的时间特征学习支路, 最终通过时空特征融合构建了一种新的认知负荷评估算法. 基于公开数据集的测试结果表明, 本文所提算法的四分类正确率达到99.27%(被试内)和95.16%(跨被试), 在现有算法中性能最优. 消融实验表明, 算法的时、空特征提取模块可分别对脑电频谱图的时空特征进行有效表征, 所提出的双路网络结构能有效完成时空特征的高效融合.
英文摘要
      Accurate cognitive workload assessment is of great significance for enhancing human-machine coordination and improving the efficiency of human-machine integration systems. Due to the low spatial resolution and the temporal fluctuation of electroencephalogram (EEG) spectrum, commonly used EEG based methods for cognitive workload assessment are not effective in utilizing spatial and temporal information among the EEG spectrum. In this work, a novel cognitive workload assessment algorithm is proposed by fusing the spatial and temporal features of the EEG spectrum, which are extracted by a CapsNet and a long short term memory network. Test results based on public datasets show that the proposed algorithm achieves the optimal performance among SOTA algorithms, reaching 99.27% (data-dependent) and 95.16% (data-independent). The ablation experiments prove that the temporal and spatial feature extraction modules of the algorithm can effectively represent the corresponding features of EEG spectrum, and the proposed dual-stream network structure can accomplish the efficient fusion of temporal and spatial features.