基于并行多尺度时间卷积网络的运动想象信号分类方法
Classification of Motor Imagery Signal Based on Parallel Multi-scale Time Convolutional Network
刘凯 1毕峰2
作者信息
- 1. 沈阳化工大学 计算机学院,辽宁 沈阳 110142;辽宁省化工过程工业智能化技术重点实验室,辽宁 沈阳 110142
- 2. 辽东学院 信息工程学院,辽宁 丹东 118003
- 折叠
摘要
运动想象信号的非线性和低时空域分辨率会导致基于深度学习分类方法的准确性较低.因此,提出一种并行多尺度时间卷积网络分类方法对运动想象信号进行分类.所提方法可提取运动想象信号在时间和空间上的信息,并在时间维度上使用并行卷积计算方法,更好地提取信号的时间特征.此外,在不影响分类准确率的前提下,使用一种简化的预处理方法,简化复杂的预处理过程.实验结果表明,所提方法在BCI Competition IV数据集上的分类准确率为 74.4%,待训练的参数量减少到 35 663;对比FBCSP和DeepNet等经典分类方法,分类准确率分别提高了 2.2%和 6.6%,参数量分别降低了 86.4%和 87.5%,验证了所提方法的有效性.
Abstract
The accuracy of classification methods based on deep learning is low because of the nonlinearity and low temporal spatial resolution of motor imagery signals.Therefore,a parallel multi-scale time convolutional net-work classification method was proposed to classify motor imagery signals.The proposed method extracted the time and space information of the motor imagery signal respectively,and uses parallel convolution computation method in the time dimension to extract the time features of the signal better.In addition,a simplified preprocessing method was used to simplify the complex preprocessing without affecting the accuracy of classification.Experimental results show that the classification accuracy of the proposed method on the BCI Competition IV dataset is 74.4%,and the number of parameters to be trained is reduced to 35 663.Compared with the classical classification methods such as FBCSP and DeepNet,the classification accuracy is increased by 2.2%and 6.6%,respectively,and the number of parameters is reduced by 86.4%and 87.5%,respectively,which verifies the effectiveness of the proposed method.
关键词
运动想象信号/卷积神经网络/脑机接口/信号分类/时空特征Key words
motor imagery signal/convolutional neural network/brain-computer interface/signal classifica-tion/spatio-temporal characteristics引用本文复制引用
基金项目
辽宁省教育厅高等学校基本科研项目(LJKMZ20221756)
出版年
2024
NETL
NSTL
万方数据