基于Gabor变换特征选择的ECG心拍分类

Electrocardiogram beat classification based on the feature selection of Gabor transform

为了解决小波变换法不能有效地提取心拍的局部特征,甚至丢失一些关键特征的问题,该文基于类内类间距离和准确性准则,分别采用顺序浮动前向搜索(SFFS)法和单独最优特征组合法对心拍信号的Gabor变换系数的实部选择。选出的系数与RR间期一起组成最近邻分类器的输入特征向量。数据集来自MIT-BI H(MassachusettesInstitute of Technology-Boston s Beth Israel Hospital)心电数据库的8种类型的心拍。实验结果表明,基于准确性准则和SFFS法即封装法的心拍分类方法准确性最高,为98.65%,仅需利用15个Gabor变换系数,有利于提高分类速度。该方法是一种更有效的心拍分类方法。

Wavelet transforms cannot effectively extract the local features of heartbeats since they lose some key features. The intra-inter distance and accuracy criterion were both used to select the real parts of the Gabor transform coefficients using a sequential floating forward selection （SFFS） algorithm and an individual optimal feature combination algorithm. The selected coefficients together with the RR interval are used as input feature vectors to a nearest neighbor classifier. Eight types of heartbeats from the MIT-BIH （Massachusetts Institute of Technology-Boston＇s Beth Israel Hospital） ECG （electrocardiogram） database are classified. Experimental results show that the approach using the intra-inter distance criterion and the SFFS strategy namely wrapper approach for heartbeat classification has the highest classification accuracy of 98.65%. The method uses only 15 coefficients and the classification speed is quite fast. This method is more effective for heartbeat classification than previous methods.