CEEMDAN与多特征融合的脑电信号识别研究  

Research on EEG Signal Recognition Based on CEEMDAN and Multi-Feature Fusion

作　　者：张聪聪 常湛源 李传江[1] ZHANG Cong-cong;CHANG Zhan-yuan;LI Chuan-jang(The College of Information,Mechanical and Electrical Engineering,Shanghai Normal University,Shanghai 200234,China;Shanghai Engineering Research Center of Intelligent Education and Bigdata,Shanghai Normal University,Shanghai 200234,China)

机构地区：[1]上海师范大学信息与机电工程学院,上海200234 [2]上海智能教育大数据工程技术研究中心,上海200234

出　　处：《计算机仿真》2022年第12期366-372,共7页Computer Simulation

摘　　要：为了解决脑电信号的非平稳特性难以分析,以及分类识别率低等问题,研究设计了基于自适应噪声完备经验模态分解(CEEMDAN)与多特征融合方法。通过对原始信号加时间窗进行CEEMDAN分解,根据各分量相关系数设计合成新信号进行共空间模式(CSP)提取空域特征,利用希尔伯特变换构造瞬时能量差和边际能量差特征,对各通道信号计算模糊熵组合成时-频-空域-非线性动力学的融合特征向量,最后采用灰狼算法(GWO)优化的支持向量机(SVM)对组合特征进行分类,对BCI Competition Ⅱ数据集平均分类识别率达89.29%。实验结果表明,多特征融合的识别率高于单一特征,加滑动时间窗改进CEEMDAN分解的方法有更高的分类识别率。In response to the non-linear and non-stationary characteristics of EEG signals and the low classification accuracy, the complete ensemble empirical mode decomposition with adaptive noise method(CEEMDAN) is introduced to extract EEG signal features. A novel feature extraction method of multi-feature fusion based on CEEMDAN decomposition improved by adding time window was proposed to improve the classification accuracy of EEG signals. Multiple intrinsic mode functions(IMFs) were obtained by CEEMDAN decomposition of the original signal with time window sliding. The IMFs are selected according to the correlation coefficients between the IMFs and the original signal, and combined into a new signal to extract spatial features by common spatial pattern(CSP). The instantaneous energy difference and marginal energy difference characteristics were constructed by using Hilbert transform, and the fuzzy entropy of each channel signal was calculated to form a fusion eigenvector of time-frequency spatial nonlinear dynamics. Fuzzy entropy was calculated for each channel signal, and the fusion eigenvector of time-frequency-space-nonlinear dynamics was combined. Gray Wolf Algorithm(GWO) optimized Support Vector Machine(SVM) was used to classify the combined features of left-handed and right-handed motor imagery with an average classification recognition rate of 89.29%. The experimental results show that recognition rate with multi-feature fusion is higher than with single feature, and the method of improving CEEMDAN decomposition by adding sliding time window has a high classification recognition rate.

关 键 词：脑电信号 自适应噪声完备经验模态分解 希尔伯特变换 模糊熵 特征融合 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]