基于WT-SVD-SVM和WT-SVD-KNN的运动想象脑电信号特征提取及分类

Feature Extraction and Classification of Motor Imagery Electroencephalogram Signals Based on WT-SVD-SVM and WT-SVD-KNN

针对运动想象脑电信号策略缺乏多样性的问题,提出一种在肢体运动的同时大脑想象运动过程与完全想象运动过程相结合的运动想象脑电信号策略。结合小波变换(wavelet transform,WT)与奇异值分解(singular value decomposition,SVD)提取运动想象脑电信号特征,用支持向量机(support vector machine,SVM)与K最近邻(K-nearest neighbor,KNN)分类器对不同运动想象策略进行分类。针对脑电信号采集过程中常出现信号失真的情况,提出了自回归(auto-regression,AR)模型结合奇异值分解的规避方法,结果表明此方法能够有效排除信号采集过程中出现的失真情况。通过比较方法WT-SVD-SVM与WT-SVD-KNN的特征提取和分类算法,结果表明,WTSVD-SVM方法在单一策略和两者组合策略中最低分类精度达到90.00%,并且该方法在想象箭头向上、箭头向右以及左拳右摆策略下精度能够达到91.11%。

Aiming at the lack of diversity in the strategy of motor imagery electroencephalogram(EEG)signals,this paper proposed a motor imagery EEG signal strategy combining the movement process imagined by brain and the complete imagination one at the moment of body movement.The features of EEG signals were extracted by wavelet transform(WT)and singular value decomposition(SVD),and classified by support vector machine(SVM)and K-nearest neighbor(KNN)classifiers.Since the signal distortion often occurs in the process of EEG signal acquisition,an auto-regressive(AR)model combined with SVD was proposed.The results show that the combination of AR coefficient and SVD can effectively eliminate the distortion.Compared to the WT-SVD-KNN method,the WT-SVD-SVM one has a minimum classification accuracy of 90.00%at single and combination motor imagery strategies.Besides,this method can achieve 91.11%accuracy at imagining the arrow up,arrow to the right,and left fist then to the right.