基于独立成分分析的优选N200和P300特征通道算法

发布时间：2018-07-08 12:26

本文选题：通道选择 + 独立成分分析　；参考：《计算机工程与科学》2017年09期

【摘要】：针对脑电信号存在个体差异性并易受噪声、伪迹干扰的特点,提出一种基于独立成分分析ICA的优选特征通道算法。采用ICA将通道的数据分解为N200、P300、眼电伪迹以及其他生理信号,根据这些信号对每个通道的影响程度,判定各通道是否适合进行特征提取。分别采用本方法和三种常用方法对12个被试的脑电数据进行特征通道选择,并进行N200和P300电位的辨识,经比对发现,本文方法取得了93.10%的平均分类准确率,比其他三种方法下的准确率分别高出7.27%、1.07%和75.96%。为预测任意被试的最优通道,采用最小二乘法对ICA权值和通道选择阈值之间的关系进行拟合,对三个新被试进行最优通道预测和电位的辨识,得到较高的分类准确率,说明此预测方法具有一定普适性。
[Abstract]:Aiming at the characteristics of individual difference, noise and artifact interference in EEG signals, an ICA based optimal feature channel selection algorithm is proposed. ICA is used to decompose the channel data into N200P300, eye electrical artifacts and other physiological signals. According to the influence of these signals on each channel, it is determined whether each channel is suitable for feature extraction. This method and three common methods were used to select the characteristic channels of EEG data of 12 subjects, and the potential of N200 and P300 were identified. By comparison, it was found that the average classification accuracy of this method was 93.10%. The accuracy was 7.27% and 75.96% higher than that of the other three methods, respectively. In order to predict the optimal channel of arbitrary subjects, the relationship between ICA weight and channel selection threshold was fitted by the least square method, and the optimal channel prediction and potential identification were carried out on three new subjects, and a higher classification accuracy was obtained. It shows that this prediction method has certain universality.
【作者单位】：天津大学电气与自动化工程学院;加州州立大学贝克斯菲尔德分校计算机系&电气工程与计算机科学学院;
【基金】：国家自然科学基金(61473207)
【分类号】：R318;TN911.6

【相似文献】