论文部分内容阅读
脑机接口系统的核心问题之一是信号分类。本文针对脑电信号的异构融合特征的分类问题提出了一种新方法:封装式弹性网特征选择和分类。首先,对预处理后的脑电(EEG)信号联合应用时域统计、功率谱、共空间模式和自回归模型方法提取高维异构融合特征。其次,采用封装方式进行特征选择:对训练数据采用弹性网罚逻辑回归拟合模型,通过坐标下降法估计模型参数,运用10倍交叉验证选择出最优特征子集。最后采用已训练的最优模型对测试样本进行分类。实验中采用国际BCI竞赛Ⅳ的EEG数据,结果表明,该方法适用于高维融合特征的最优特征子集选择问题,对于EEG信号的识别不仅效果好、速度快,而且能够选出与分类更相关的子集,获得相对简单的模型,平均测试正确率达到了81.78%。
One of the core issues of the BCI system is signal classification. In this paper, we propose a new method for the classification of heterogeneous fusion features of EEG: the feature selection and classification of encapsulated elastic network. Firstly, high-dimensional heterogeneous fusion features were extracted from time-domain statistics, power spectrum, co-space model and autoregressive model for preprocessed EEG signals. Secondly, the feature selection is made by encapsulation method: the elastic net penalty logistic regression fitting model is used for the training data, the model parameters are estimated by the coordinate descent method, and the optimal feature subset is selected by 10 times cross-validation. Finally, the best training model has been used to classify the test samples. The results show that the proposed method is suitable for the selection of the optimal feature subsets of high dimensional fusion features. The recognition of EEG signals is not only effective and fast, but also can be used to select more Relative subset, get a relatively simple model, the average test accuracy rate reached 81.78%.