EEG信号识别中可调优化Q因子小波变换的多特征融合算法

Multi-feature fusion algorithm for adaptive-tunable Q-factor wavelet transform in EEG signal recognition

EEG信号对脑部疾病诊断具有重要意义,但其特征选择对信号识别准确率影响较大.针对这个问题,本文基于Q因子小波变换,提出一种可调优化Q因子小波变换融合多维特征的脑电信号识别(Ad-TQWT MF)算法.该算法首先根据小波分解后的子带信号定义能量香农熵比,用其作为可调Q因子小波的优化评价标准;再融合变换后信号的时域,频域和非线性特征,通过自适应特征选择方法构建特征子空间;最后在该特征子空间下对脑电信号进行识别.在BCI脑电竞赛数据集DatasetⅢ,O3VR,X11b和S4b进行了实验,实验结果表明:Ad-TQWT MF算法在LDA分类器中精度为89.2%,81.2%,83.2%和85.6%,相比于原Q因子小波变换,冗余特征减少10%~30%,相较于Haar和Db 4小波精度提高3%~5%,证明了Ad-TQWT MF算法的有效性.

EEG signals are of great significance for the diagnosis of brain diseases.However,the recognition accuracy is greatly affected by the feature selection and redundant features.To solve this problem,an adaptive-tunable Q-factor wavelet transform multi-feature algorithm(Ad-TQWT MF)is proposed based on the Q-factor wavelet transform in this paper.First,the energy Shannon entropy ratio based on the decomposed subbands is defined,and it is set to the optimization evaluation standard of the adjustable Q factor wavelet.Then,the time domain,frequency domain and nonlinear features of the transformed EEG are fused to construct a useful characteristic subspace through the adaptive feature selection.Finally,the characteristic subspace of EEG is identified by the Ad-TQWT MF.Some experiments are carried out on the EEG datasets of first and second BCI competitions including DatasetⅢ,O3VR,X11b and S4b.The experimental results show that the accuracy of the proposed algorithm in the linear discriminant analysis classifier are 89.2%,81.2%,83.2%and 85.6%.Compared to Haar and Db 4,the feature redundancy rate of proposed algorithm drops by 10%~30%and the recognition accuracy of EEG increases 3%~5%.The experimental results strongly prove the effectiveness of the Ad-TQWT algorithm.