基于改进注意力机制的领域对抗网络的认知负荷识别模型  

Cognitive Load Recognition Model Based on Improved Attention Mechanism in Domain-adversarial Networks

作　　者：班瑞阳 周大鹏 韩吉平 刘文海 BAN Ruiyang;ZHOU Dapeng;HAN Jiping;LIU Wenhai(China Aviation Industry Corporation Shenyang Aircraft Design and Research Institute,Shenyang 110066,China;Shenyang Institute of Computing Technology,Chinese Academy of Sciences,Shenyang 110168,China)

机构地区：[1]中国航空工业集团公司沈阳飞机设计研究所,沈阳110066 [2]中国科学院沈阳计算技术研究所,沈阳110168

出　　处：《小型微型计算机系统》2024年第11期2602-2608,共7页Journal of Chinese Computer Systems

基　　金：国家2023年产业基础再造和制造业高质量发展专项项目(TC230A076-13)资助.

摘　　要：在认知负荷识别领域,精确的跨领域识别对提高模型的鲁棒性和适应性至关重要,其中基于脑电信号(EEG)的评价方法当前已经成为研究的主流方向,但是由于脑电信号自身具有差异性非稳态性的特点,因此需要提高脑电信号在测量时的泛化性,有效实现通过测量EEG信号进行认知负荷识别.本文提出了一种增加注意力机制的长短时记忆网络和领域对抗网络相结合的深度学习模型,在领域对抗网络中加入了通过经验模态分解(EMD)计算的近似熵注意力机制,该模型通过集成注意力机制增强特征提取能力,能够有效捕捉与认知负荷相关的关键信息;同时通过源域和目标域之间的不断对抗,混淆源域与目标域之间的分布差异,达到提高模型识别泛化性的效果.在完成模型构建之后,选择现有方法与本文提出模型进行对比,取得了较好成绩,证明了本文模型在EEG识别中的优越性.In the field of cognitive load recognition,accurate cross-domain recognition is crucial to improve the robustness and adaptability of the model,in which the evaluation method based on electroencephalographic(EEG)signals has now become the mainstream direction of research,but due to the EEG signals themselves have the characteristics of the differential non-stationary,so it is necessary to improve the EEG signals at the time of the measurement of the generality of the EEG signals,and effectively achieve the cognitive load through the measurement of EEG signal Recognition.In this paper,we propose a deep learning model that combines long and short-term memory network and domain adversarial network by adding an attention mechanism,and an approximate entropy attention mechanism computed by empirical modal decomposition(EMD)is added to the domain adversarial network,which enhances the feature extraction ability by integrating the attention mechanism,and is able to effectively capture the key information related to cognitive load;at the same time,through constant adversarial confrontation between the source and target domains,it confuses the source domain and the target domain,and the source and target domains can be confused with the target domains.constant confrontation,the distribution difference between the source domain and the target domain is confused to achieve the effect of improving the model recognition generalisation.After completing the model construction,the existing methods are selected and compared with the proposed model in this paper,and better results are achieved,proving the superiority of this paper′s model in EEG recognition.

关 键 词：长短时记忆网络 领域对抗网络 注意力机制 认知负荷识别 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]