多尺度卷积与注意力机制的脑电信号识别研究  

Research on EEG signal recognition based on multi⁃scale convolutionand attention mechanism

作　　者：冯泽林 宋耀莲[1] FENG Zelin;SONG Yaolian(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500

出　　处：《现代电子技术》2023年第23期85-92,共8页Modern Electronics Technique

摘　　要：基于运动想象的脑机接口技术可以使大脑绕过中枢和外周神经,在人脑和外部电子设备之间构建直连通路,帮助运动受限人群直接与外界进行交互。由于运动想象脑电信号具有信噪比和空间分辨率较低的特性,导致解码效率较低,文中提出一种基于注意力机制的解码运动想象脑电信号的深度学习模型MSATCNet。首先原始脑电信号通过并行多尺度卷积神经网络提取全局特征和局部特征并加以融合,通过多头注意力机制模块突出融合后的重要特征,最后由时间卷积网络(TCN)提取时序信息,全连接层和Softmax层对提取后的特征进行学习和分类。对所提出的模型在BCI竞赛Ⅳ-2a数据集上进行了实验与分析,所提出模型对所有受试者平均分类精度达到了83.99%,其中最高准确率达到97.07%。结果表明,所提出的模型可以有效提高MI-EEG的分类准确率,提升了运动想象脑电解码的可靠性。The brain⁃computer interface technology based on motor imagery(MI)can make the brain bypass the central and peripheral nerves.In the technology,a direct connection between the human brain and external electronic devices is established,which helps people with limited movement to interact with the external world directly.The decoding efficiency is low because of the low signal⁃to⁃noise ratio(SNR)and spatial resolution of motor imagery EEG(MI⁃EEG)signals.Therefore,a deep learning model MSATCNet based on attention mechanism is proposed to decode MI⁃EEG signals.Global and local features of the original EEG signals are extracted by a parallel multi⁃scale convolutional neural network(CNN)and fused.The important features after fusion are highlighted by the multi⁃head attention mechanism module.The time series information is extracted by a temporal convolutional network(TCN).The fully connection layer and the Softmax layer learn and classify the extracted features.The proposed model is tested and analyzed on the dataset BCI competitionⅣ⁃2a.The average classification accuracy of the proposed model for all subjects reached 83.99%,and its highest accuracy reached 97.07%.The experimental results indicate that the proposed model can improve the classification accuracy of MI⁃EEG effectively and improve the reliability of MI⁃EEG decoding.

关 键 词：运动想象 多尺度卷积 注意力机制 脑机接口 时间卷积网络 脑电信号识别 

分 类 号：TN911.7-34[电子电信—通信与信息系统] TP391.9[电子电信—信息与通信工程]