摘要
针对传统深度学习方法在解码脑电信号时可能存在的特征挖掘不足及利用不充分问题,提出一种并行多尺度时间卷积结合滑动窗口技术与注意力机制的深度学习模型,即MCSANet。首先,利用并行多尺度时间卷积有效捕获脑电信号在不同时间尺度下的时域特征和空域特征;再利用滑动窗口切片技术对特征序列进行划分,增加特征序列样本数;之后,每部分特征序列样本都通过多头自注意力机制分配权重并加以融合,进一步突显出更多关键特征;最后,全连接层和SoftMax层共同协作,对捕获到的特征进行深入学习和精准分类。为了验证该模型的性能,在BCICIV-2a数据集上进行了详尽的实验分析。结果表明,所有受试者的平均分类准确率都高达81.69%,验证了所提出的方法在挖掘脑电深层潜在特征、提升运动想象脑电分类性能方面的有效性。
In order to solve the problem of insufficient feature mining and insufficient utilization in decoding electroencephalography(EEG)signals by means of the traditional deep learning method,a deep learning model,MCSANet,is proposed,which combines the parallel multi-scale temporal convolution with sliding window technology and attention mechanism.The parallel multi-scale temporal convolution is used to effectively capture the temporal characteristics and spatial characteristics of EEG signals at different time scales.The sliding window slicing technology is used to divide the feature sequences and increase the number of sequence samples.The weights of each part of the sequence samples are assigned and fused by means of the multi-head self-attention mechanism,which can further highlight more key features.The fully connected layer and the SoftMax layer are used to work together,so as to perform in-depth learning and accurate classification for the captured features.In order to validate the performance of the model,an exhaustive experimental analysis was performed on the BCICIV-2a dataset.The experimental results show that the average classification accuracy of all subjects is as high as 81.69%,which verifies the effectiveness of the proposed method in mining the deep potential features of EEG and improving the classification performance of motor imagery EEG.
作者
杜江
毕峰
DU Jiang;BI Feng(School of Computer Science and Technology,Shenyang University of Chemical Technology,Shenyang 110142,China;Liaoning Provincial Key Laboratory of Intelligent Technology for Chemical Process Industry,Shenyang 110142,China;School of Information Engineering,Liaodong University,Dandong 118003,China)
出处
《现代电子技术》
北大核心
2025年第16期67-74,共8页
Modern Electronics Technique
基金
辽宁省教育厅高校基本科研项目(LJKMZ20221756)
辽宁省科技计划联合计划项目:脑机接口系统中特征脑电信号的提取与分类关键技术研究(2024JH2/102600133)。
关键词
脑机接口
脑电信号
并行多尺度时间卷积
滑动窗口切片技术
多头自注意力机制
消融实验
braincomputer interface
EEG signal
parallel multi-scale temporal convolution
sliding window slicing technology
multi-head self-attention mechanism
ablation experiment
