模拟阅读型脑-机接口信号分类研究  

Research of signal classification for brain-computer interface

作　　者：朱学才[1] 李梅[2] 邹思轩[3] 

机构地区：[1]九江职业技术学院继续教育部,江西九江332007 [2]中南民族大学生物医学工程学院,武汉430074 [3]华中科技大学电子与信息工程系,武汉430074

出　　处：《华中师范大学学报（自然科学版）》2011年第3期391-395,共5页Journal of Central China Normal University：Natural Sciences

基　　金：国家自然科学基金资助项目(30370393)

摘　　要：脑-机接口(BCI)研究中的一个关键问题是如何正确地对EEG信号进行模式分类,以输出控制命令.本文在对"模拟自然阅读"模式下非靶刺激和靶刺激诱发的EEG进行去均值、低通滤波、下采样和归一化等处理后,结合对视觉诱发事件相关电位时域特征分析,提取出最佳特征量,分别利用BP神经网络和线性感知器算法对这些特征模式进行了分类.最终平均识别正确率分别达到87%和84%以上.对比研究表明,BP神经网络算法的分类效果较好,推测这是由于大部分EEG模式线性可分,只有10%左右线性不可分但非线性可分造成的.为提高分类正确率,简化BCI设计,详细研究了信号时程、时段的选择以及通道的选取对模式分类精度的影响.结果表明,信号时程越长分类精度越高;信号时段的选择对分类精度亦有较大的影响.通过实验发现:32个通道中,选取第14(PO3)通道的EEG进行模式分类的精度最高.In order to produce the output of command, a key issue in brain-computer interface （BCI） is to classify EEG signals correctly. EEG signal preprocessing were dis- cussed in this paper, which include the lowpass filtering, baseline removing, down-sampling and normalization, et al. The EEG signals were recorded in an ＂Imitating Nature Reading＂ modality. The optimal feature patterns were extracted basing on the analyses in temporal and frequency for the Visual Evoked Event Related Potentials, and then BP Neural Networks and the perceptron approach were used separately to classify these pat- terns. The best classification results on testing set revealed an accuracy of more than 87~ and 84 ~ respectively. The effects of classification by BP Neural Networks were better than those by perceptron approach, which revealed that about 90 ~ EEG patterns are linearly separable while other 10o//00 are linearly inseparable but maybe non-linearly separable. In order to get a better accuracy of classification, we investigated how the classification accuracy was affected by the selection of signal intervals. The results show that the longer the length of signals, the better the accuracy of classification. The selection of signal intervals also played a key role in classification. The works are essential for boosting up the speed of whole BCI system. Overpass the experiment, we can find that the accuracy of classification is the highest when we take the 14th channel standard to classify the patterns.

关 键 词：脑-机接口 模式识别 BP神经网络 感知器算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]