融合注意力的双分支时空卷积脑电识别网络

打开文本图片集

中图分类号：TN911.7-34 文献标识码：A 文章编号：1004-373X（2025）18-0159-06

DOI：10.16652/j.issn.1004-373x.2025.18.024

引用格式：，，.融合注意力的双分支时空卷积脑电识别网络[J].现代电子技术，2025，48（18）：159-164.

Abstract：In alusionto the problemsof weak feature expressionof motor image（MI）electroencephalography （EEG）signals andlowauracyofclassifiationandrecogitionadual-banchspatitempralconvolutionalmotorimageryrecogitionetwork integratingatentionmechanismisproposed.Inordertoaddress theproblemoffewdata，animprovedslidingwindowtechology isproposedforthedataaugmentation.Thedataisfedintothefeatureextractionnetwork，andthedual-branchstructureisused tosimultaneouslyextractthefeatureinboth temporalandspatialaspects.Onthetemporalbranch，multi-scaletemporal convolutionisusedtoextracttempralfeaturesatdiferentscales.Onthespatialbranch，depspatialfeaturesareusedtoextract depth-separableconvolution.Theatentionmechanismisused todynamicallyassignweightstothefusedfeatures.The comparison resultswith other methodsshow that theproposednetwork model hasbeter classfication performance and generalization ability.

Keywords：EEGsignal;motorimagery；multi-scale temporalconvolution;dual-branchstructure；dataaugmentation; attention mechanism

0 引言

脑机接口（Brain-ComputerInterface，BCI）系统可以将脑电信号解码成计算机设备可识别的指令，进而实现大脑活动与外部设备的交互-2]。（剩余8905字）