【摘要】：腦機(jī)接口系統(tǒng)的核心問題之一是信號(hào)分類。本文針對(duì)腦電信號(hào)的異構(gòu)融合特征的分類問題提出了一種新方法:封裝式彈性網(wǎng)特征選擇和分類。首先,對(duì)預(yù)處理后的腦電(EEG)信號(hào)聯(lián)合應(yīng)用時(shí)域統(tǒng)計(jì)、功率譜、共空間模式和自回歸模型方法提取高維異構(gòu)融合特征。其次,采用封裝方式進(jìn)行特征選擇:對(duì)訓(xùn)練數(shù)據(jù)采用彈性網(wǎng)罰邏輯回歸擬合模型,通過坐標(biāo)下降法估計(jì)模型參數(shù),運(yùn)用10倍交叉驗(yàn)證選擇出最優(yōu)特征子集。最后采用已訓(xùn)練的最優(yōu)模型對(duì)測(cè)試樣本進(jìn)行分類。實(shí)驗(yàn)中采用國(guó)際BCI競(jìng)賽Ⅳ的EEG數(shù)據(jù),結(jié)果表明,該方法適用于高維融合特征的最優(yōu)特征子集選擇問題,對(duì)于EEG信號(hào)的識(shí)別不僅效果好、速度快,而且能夠選出與分類更相關(guān)的子集,獲得相對(duì)簡(jiǎn)單的模型,平均測(cè)試正確率達(dá)到了81.78%。
[Abstract]:One of the core problems of BCI system is signal classification. In this paper, a new method for the classification of heterogeneous fusion features of EEG signals is proposed: feature selection and classification of encapsulated elastic networks. Firstly, time domain statistics, power spectrum, common space model and autoregressive model are used to extract the features of high dimensional heterogeneous fusion for pretreated (EEG) signals. Secondly, the feature selection is carried out by encapsulation: the training data is fitted with elastic net penalty logic regression model, the parameters of the model are estimated by coordinate descent method, and the optimal feature subset is selected by 10 times cross validation. Finally, the trained optimal model is used to classify the test samples. In the experiment, the EEG data of international BCI competition 鈪，

本文編號(hào)：2363452