一种多尺度特征融合运动想象脑电信号分类算法——基于时频域数据增强  

A Classification Algorithm for Multi-scale Feature Fusion Motor Imagery EEG Signals——Based on Time-frequency Domain Data Augmentation

作　　者：郑敏敏[1] 钱政康 赵桐 ZHENG Minmin;QIAN Zhengkang;ZHAO Tong(School of Mechanical,Electrical and Information Engineering,Putian University,Putian Fujian 351131,China)

机构地区：[1]莆田学院机电与信息工程学院,福建莆田351131

出　　处：《莆田学院学报》2025年第2期52-60,70,共10页Journal of putian University

基　　金：福建省科技厅面上项目(2021J011108);莆田学院科研启动项目(2023058);莆田学院研究生科研创新项目(科技类)(yjs2024057)。

摘　　要：针对传统的脑电信号识别算法样本量较小、识别分类准确率较低的问题,提出了一种基于时频域数据增强的多尺度特征融合运动想象脑电信号分类算法。对原始信号进行独立主成分分析滤除眼电噪声,并进行时域和频域的数据增强,将原始数据、时域增强数据、频域增强数据三组数据分别进行多尺度多频带的共空间模式、功率谱密度以及小波包能量特征提取,合并特征并通过特征选择得到最佳特征组,训练支持向量机,由投票得出最终分类结果。在公开数据集BCI CompetitionⅣ-2a和OpenBMI上进行验证,结果表明提出的方法分类效果良好,分类准确率高于其他对比方法。Aiming at the small sample size and low classification accuracy of traditional EEG recognition algorithms,a motor imagery EEG classification algorithm based on multi-scale feature fusion and time-frequency domain data augmentation is proposed.Independent component analysis(ICA)is used to remove ocular artifacts from raw signals,followed by data augmentation in the time and frequency domains.Three types of data——original,time-augmented,and frequency-augmented undergo multi-scale,multi-frequency-band feature extraction,including common spatial pattern(CSP),power spectral density(PSD),and wavelet packet energy(WPE)features.The extracted features are merged and the optimal feature set is selected through feature selection.Then,a support vector machine(SVM)classifier is trained and the final classification result is determined by a voting mechanism.Experiments on the BCI CompetitionⅣ-2a and OpenBMI datasets show that the proposed method achieves higher classification accuracy than other comparative methods.

关 键 词：运动想象 数据增强 共空间模式 支持向量机 

分 类 号：TN911.7[电子电信—通信与信息系统]