基于BP神经网络的运动想象脑电信号分类研究

运动想象脑机接口因具有更大的自主性、灵活性，在脑机互联领域得到了广泛应用，相比较其它范式分类准确率偏低，限制了其发展。本文利用时频图谱、脑地形图两种特征分析方法对上肢运动想象脑电信号进行了特征分析，并采用滤波器组共空间模式(filter bank co-space，FBCSP)特征提取算法对上肢运动想象信号数据进行了特征提取，再将提取结果分别利用支持向量机（support vector machine，SVM）算法、K-最近邻(K-Nearest Neighbor)算法、反向传播（back propagation，BP）神经网络三种分类算法进行分类，研究结果发现SVM算法、KNN算法、BP神经网络算法应用在上肢运动想象脑机接口系统的平均分类准确率分别为76.45%、74.55%、81.70%，BP神经网络算法相比SVM算法、KNN算法在上肢运动想象任务的分类准确率上分别高出了5.25%、7.15%，并且t检验后得到分类准确率均具有极显著的统计学差异，并利用ROC曲线和AUC值检测了分类器效果，BP神经网络的AUC值相比SVM算法、KNN算法也分别提升了0.1226、0.1285，表明BP神经网络分类算法相比较SVM算法、KNN算法更适用于上肢运动想象脑机接口系统，提高了系统的分类准确率，推动了上肢运动想象脑电信号实际应用的发展进程。

Research on EEG signal classification of motor imagery based on BP neural network

The motor imagery brain-computer interface has been widely used in the field of brain-computer interconnection due to its greater autonomy and flexibility. Compared with other paradigms, the classification accuracy rate is low, which limits its development. In this paper, two feature analysis methods of time-frequency atlas and brain topography were used to analyze the EEG signals of upper limb motor imagery, and the filter bank co-space (FBCSP) feature extraction algorithm was used to analyze the characteristics of upper limb motor imagery. The signal data is extracted with features, and then the extraction results are divided into three types: support vector machine (SVM) algorithm, K-Nearest Neighbor (K-Nearest Neighbor) algorithm, and back propagation (BP) neural network. According to the classification algorithm, the research results found that the average classification accuracy of the SVM algorithm, KNN algorithm and BP neural network algorithm applied to the upper limb motor imagery brain-computer interface system was 76.45%, 74.55%, and 81.70%, respectively. Compared with the SVM, the BP neural network algorithm The algorithm and KNN algorithm are 5.25% and 7.15% higher in the classification accuracy of the upper limb motor imagery task respectively, and the classification accuracy obtained after the t test has a very significant statistical difference, and the ROC curve and AUC value were used to detect Compared with the SVM algorithm and the KNN algorithm, the AUC value of the BP neural network is also increased by 0.1226 and 0.1285 respectively, indicating that the BP neural network classification algorithm is more suitable for the upper limb motor imagery brain-computer interface system than the SVM algorithm and the KNN algorithm. , improve the classification accuracy of the system, and promote the development process of the practical application of upper limb motor imagery EEG signals.