基于回溯搜索算法的导联选择脑机接口研究

摘要

在脑机接口（BCI）中，传统的共空域模式（CSP）算法在提取特征信号与事件相关去同步/同步（ERD/ERS）的信息上得到了很好的效果。但是CSP算法受限于电极导联数、EEG信号的时间段和频带等因素，如电极导联数的增加，CSP算法容易过拟合，数据记录容易混乱，使得运算变得复杂，增加运算时间，降低数据分类正确率。所以，CSP算法存在局限性。使用回溯搜索优化算法（BSA）能够为CSP算法自动挑选出一组导联数组子集，并且以分类错误率作为BSA算法的目标函数进行实验。实验采用两类实验数据（第三、四届国际BCI竞赛数据集）进行交叉验证分类实验。实验结果表明，两类数据的导联数目大幅度减少，分类正确率有所提高。%In brain⁃computer interface(BCI),the traditional common spatial pattern(CSP)algorithm has a good effect on characteristic signal extraction and event⁃related desynchronization/event⁃related synchronization (ERD/ERS) information. The CSP algorithm is easily limited by electrode lead quantity,time period and frequency band of EEG signal,such as the increasing of electrode lead quantity,easy overfitting of CSP algorithm and easy chaos of data record,which can make the operation com⁃plex,increase the operation time and reduce the accuracy of data classification. Therefore,the CSP algorithm have a limitation. The backtracking search optimization algorithm(BSA)proposed in this paper can automatically select a subset of lead array for CSP algorithm,and take the classification error rate as the objective function of BSA algorithm to test. The two datasets from the datasets of the Third,Fourth International BCI Competitions are adopted in the experiment to perform the classification experi⁃ment of cross validation. The experimental results show that the lead quantity of the two datasets are dramatically reduced,and the classification accuracy is improved.