基于CSP与卷积神经网络算法的多类运动想象脑电信号分类
发布时间：2018-07-14 14:56
【摘要】：针对直接利用卷积神经网络(convolutional neural network,CNN)算法对多类运动想象脑电信号分类识别时,因样本量比较少,难以充分训练权值,导致分类效果较差的问题,结合一对多CSP算法与CNN算法对多类运动想象脑电信号进行特征提取与分类。首先,利用CSP算法对多类运动想象脑电信号进行特征提取,形成一维特征数据,作为CNN的输入样本;其次,对传统二维输入样本的CNN结构进行改造,使其适应一维数据的输入样本,对输入样本进行再次特征提取并分类;最后,使用BCI2005desc—Ⅲa的K3b数据进行算法验证;并对不同参数值的确定进行了讨论。算法验证结果表明,单独利用一对多CSP算法得到的分类正确率73%,单独使用CNN算法得到正确率为75%,新算法取得了91.46%的正确率,相比两种原始方法有较大提升。
[Abstract]:In order to solve the problem of classifying and recognizing multiple motion imaginary EEG signals directly using convolutional neural network (convolutional neural network) algorithm, it is difficult to fully train the weights due to the small sample size, which leads to the poor classification effect. Combined with one-to-many CSP algorithm and CNN algorithm, the feature extraction and classification of multi-class motion imaginary EEG signals are carried out. Firstly, we use CSP algorithm to extract the features of multi-class motion imaginary EEG signals, and form one-dimensional feature data as the input samples of CNN. Secondly, we modify the CNN structure of traditional two-dimensional input samples. It adapts to the input samples of one-dimensional data, extracts and classifies the input samples again. Finally, the algorithm is verified by using the K3b data of BCI2005desc- 鈪，
本文编号：2122010
[Abstract]:In order to solve the problem of classifying and recognizing multiple motion imaginary EEG signals directly using convolutional neural network (convolutional neural network) algorithm, it is difficult to fully train the weights due to the small sample size, which leads to the poor classification effect. Combined with one-to-many CSP algorithm and CNN algorithm, the feature extraction and classification of multi-class motion imaginary EEG signals are carried out. Firstly, we use CSP algorithm to extract the features of multi-class motion imaginary EEG signals, and form one-dimensional feature data as the input samples of CNN. Secondly, we modify the CNN structure of traditional two-dimensional input samples. It adapts to the input samples of one-dimensional data, extracts and classifies the input samples again. Finally, the algorithm is verified by using the K3b data of BCI2005desc- 鈪，
本文编号：2122010
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2122010.html
