基于混合脑机接口通道选择与分层特征融合的认知工作负荷识别
Cognitive Workload Recognition Based on Hybrid Brain-Computer Interface Channel Selection and Hierarchical Feature Fusion
张恒千 1詹志远 1尹钟1
作者信息
- 1. 上海理工大学 光电信息与计算机工程学院,上海 200093
- 折叠
摘要
基于脑电信号(EEG)与功能近红外光谱(fNIRS)生理数据的认知工作负荷识别研究在脑机接口领域备受关注.然而,复杂的数据采集环境对通道间数据产生不可控影响,严重制约了模拟人类大脑信息传输过程的模型准确性与完整性.针对此问题,提出一种改进的动态图注意力通道选择方法,通过图注意力网络(GAT)返回的注意力得分进行通道选择,从而减少环境干扰,提升模型鲁棒性.此外,简单特征融合会忽略不同模态间的异构性而导致重要信息丢失,因而设计了分层特征融合模块.该模块通过有效整合EEG与fNIRS不同层次的特征信息,从而增强了模型对认知任务的识别能力.在柏林工业大学提供的两个公开的心算任务和N-Back任务数据集上进行了验证,在被试依赖的训练策略下,对每名被试采用十折交叉验证方法,分别取得了85.44%和91.72%的平均精度,相比于当前先进方法,该方法表现出一定优势.实验结果表明,该模型能够在复杂数据环境下有效识别认知工作负荷,同时所提出的通道选择方法对于降低计算成本、排除无关通道具有重要意义.
Abstract
Research on the recognition of cognitive workload based on electroencephalogram(EEG)and functional near-infrared spectrosco-py(fNIRS)physiological data has garnered significant attention in the field of brain-computer interfaces.However,the complex data acquisi-tion environment introduces uncontrollable effects on inter-channel data,severely limiting the accuracy and integrity of models simulating hu-man brain information transmission processes.Therefore,this paper proposes an improved dynamic graph attention-based channel selection method.The method utilizes attention scores returned by a Graph Attention Network(GAT)to select channels,thereby reducing environmen-tal interference and enhancing model robustness.Moreover,simple feature fusion can overlook the heterogeneity between different modalities,leading to the loss of critical information.To mitigate this,we designed a hierarchical feature fusion module.We validated our approach using two publicly available datasets provided by the Berlin Institute of Technology:a mental arithmetic task and an N-Back task.Employing a sub-ject-dependent training strategy and ten-fold cross-validation for each participant,our method achieved average accuracies of 85.44%and 91.72%,respectively.Compared to current state-of-the-art methods,our approach demonstrates certain advantages.The experimental results indicate that the proposed model effectively recognizes cognitive workload in complex data environments.Additionally,the proposed channel selection method is significant for reducing computational cost and eliminating irrelevant channels.
关键词
脑电图/功能性近红外光谱/通道选择/认知工作负荷识别/分层特征融合Key words
electroencephalogram/functional near-infrared spectroscopy/channel selection/cognitive workload relognition/hierarchical feature fusion引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据