摘要
基于脑电信号的情绪识别,因其可以客观地反映人的生理和心理状态而成为了情绪调节干预的医疗辅助。针对现有方法因忽略个体间通道数据分布差异导致的情绪识别泛化性能差的问题,提出一种基于缩放卷积注意力网络的跨多个体情绪识别新方法。该方法在提取多通道脑电信号中情绪量化特征的基础上,构造新型缩放卷积注意力网络以建立不同通道、不同尺度情绪特征的协同变化关系,通过模型训练自动学习协同关系的权重,最终获得对情绪极性的域不变表征,以提高跨多个体脑电情绪识别的泛化性能。使用情绪脑电图数据集SEED和SEED-IV中的100665和100950个脑电样本进行跨多个体情绪识别。该方法在情绪三分类会和四分类中识别准确率分别为89.63%和75.65%,特别是在个体数变化情况下,其鲁棒性优于现有大多数模型。所提出的方法可有效提取情绪极性的域不变表征.
Emotion recognition based on EEG signals has become a medical aid to emotion regulation intervention because it can objectively reflect human physiological and psychological states.To address the problem of poor generalization performance of emotion recognition caused by existing methods ignoring the differences in channel data distribution between individuals,a new cross-subjects emotion recognition method based on scaled convolutional attention network was proposed in this work.Based on the extraction of emotion quantification features in multi-channel electroencephalography(EEG)signals,a novel scaled convolutional attention network was constructed to establish the synergistic change relationship of emotional features of different channels and scales,and the weight of the synergistic relationship was automatically learned by model training,and finally the domain invariant representation of the emotion polarity was obtained,which improved the generalization performance of emotion recognition across multiple individuals EEG.The emotion EEG dataset SEED and SEED-IV were used to identify emotions cross-subjects with 100665 and 100950 EEG samples,and the recognition accuracy of the proposed method was 89.63%and 75.65%in the three and four-class of emotions.Particularly,the robustness of the proposed model outperformed most of the existing methods when there were changes in the number of individuals.The results showed that the proposed method was able to effectively extract the domain-invariant representation of emotional polarity.
作者
陈彬滨
吴涛
陈黎飞
Chen Binbin;Wu Tao;Chen Lifei(School of Computer and Cyberspace Security,Fujian Normal University,Fuzhou 350117,China;Digital Fujian Environmental Monitoring Internet of Things Laboratory,Fujian Normal University,Fuzhou 35117,China;Fujian Provincial Center of Applied Mathematics,Fujian Normal University,Fuzhou 350117,China)
出处
《中国生物医学工程学报》
CAS
CSCD
北大核心
2024年第5期550-560,共11页
Chinese Journal of Biomedical Engineering
基金
国家自然科学基金(U1805263)
国家重点研发计划项目(2021YFF1200700,2021YFF1200800)。
关键词
脑电信号
情绪识别
缩放卷积网络
通道分布差异
多通道
EEG
emotion recognition
scaled convolutional network
channel distribution dfferences
multi-channel
作者简介
通信作者:陈黎飞,E-mailclfei@fjnu.edu.cn。
