基于多特征融合的多分类运动想象脑电信号识别研究

针对目前多分类运动想象脑电识别存在特征提取单一、分类准确率低等问题,提出一种多特征融合的四分类运动想象脑电识别方法来提高识别率。对预处理后的脑电信号分别使用希尔伯特-黄变换、一对多共空间模式、近似熵、模糊熵、样本熵提取结合时频—空域—非线性动力学的初始特征向量,用主成分分析降维,最后使用粒子群优化支持向量机分类。该算法通过对国际标准数据集BCI2005 Data set IIIa中的k3b受试者数据经MATLAB仿真处理后获得93.30%的识别率,均高于单一特征和其它组合特征下的识别率。分别对四名实验者实验采集运动想象脑电数据,使用本研究提出的方法处理获得了72.96%的平均识别率。结果表明多特征融合的特征提取方法能更好的表征运动想象脑电信号,使用粒子群支持向量机可取得较高的识别准确率,为人脑的认知活动提供了一种新的识别方法。

Multi-classified motion imagery of EEG signal recognition based on multi-feature fusion

To solve the problems of single feature classification and low classification accuracy of multi-classified motor imagery EEG recognition, a four-class motor imagery EEG recognition method was proposed based on multi-feature fusion to improve the recognition rate. Preprocessed EEG signals were treated by Hilbert-Huang transform, one-to-multiple common spatial pattern, approximate entropy, fuzzy entropy, and sample entropy extraction combined with the initial eigenvectors of time-frequency-space-linear dynamics. Principal component analysis was to reduce dimensionality, and particle swarm optimization was adopted to support vector machine classification. Through MATLAB simulation on the international standard data set BCI2005 Data set IIIa, the recognition rate of the algorithm reached 93.3%, which was higher than these of single feature and other combination features.The motor imagery EEG data of four experimenters was collected, and the average recognition was 72.96% by using the method proposed in this study. Results show that the feature extraction method based on multi-feature fusion can better characterize motor imagery EEG signals. Using particle swarm support vector machine can achieve higher recognition accuracy, which provides a new recognition method for human brain cognitive activities and a new recognition method for human brain cognitive activities.