摘要
针对当前脑电(EEG)情感识别技术常受冗余信号干扰的问题,提出了基于多通道帧级筛选长短时记忆网络模型(multi-channel fame-level filtered long short-term memory,MCFL-LSTM)。设计了“多头门控”模块,该模块以拼接的32个通道特征片段作为输入,通过多头机制,即采用多个门控单元获取单通道的帧级片段的特征权重,筛选出单通道中重要特征,减少冗余和无意义特征片段的影响。在帧级特征提取后,将维度变换后的32个通道输入门控单元进行通道级筛选,从而获取与当前情感刺激最相关通道,提升模型特征提取能力,增强了识别性能。实验结果表明,方法在DEAP数据集上4个二元分类评估分别达到了87.21%、82.26%、82.98%和87.53%的平均准确度,证明了模型的有效性和鲁棒性。
The"multi-channel frame-level filtered long short-term memory(MCFL-LSTM)network model"is proposed to address the issue of redundant signal interference commonly encountered in current electroencephalogram(EEG)emotion recognition techniques.The"Multi-Headed Gate"module is designed in this study.This module takes concatenated 32-channel feature segments as input and employs multiple gate units through a multi-head mechanism.The multi-head mechanism utilizes multiple gate units to obtain feature weights for individual channel-level frame segments,filtering out important features within each channel and reducing the impact of redundant and irrelevant feature segments.After frame-level feature extraction,the 32-channel inputs are dimensionally transformed and fed into the gate units for channel-level filtering.This process identifies the most relevant channels to the current emotional stimuli,thereby enhancing the model's feature extraction capability and improving recognition performance.The experimental results demonstrate the effectiveness and robustness of the proposed method on the DEAP dataset,achieving average accuracies of 87.21%,82.26%,82.98%and 87.53%for the four binary classification targets,respectively.This substantiates the effectiveness and robustness of the model.
作者
闫舒羽
李小光
顾天昊
徐冠华
Yan Shuyu;Li Xiaoguang;Gu Tianhao;Xu Guanhua(School of Automation,Qingdao University,Qingdao 260000,China;Institute of Intelligent Unmanned System,Qingdao University,Qingdao 260000,China)
出处
《国外电子测量技术》
北大核心
2023年第12期94-101,共8页
Foreign Electronic Measurement Technology
基金
国家自然科学基金(62076094)
中国博士后科学基金(2022M721744)
山东省博士后创新人才支持计划(SDBX2022023)项目资助。
关键词
EEG情感识别
帧级特征筛选
通道级筛选
神经网络
受试者无关
EEG emotion recognition
frame-level filtering
channel-level filtering
neural networks
subject-independent
