基于复合特征和FOAGRNN的心电图分类

The Electrocardiogram Classification Approach Based on Wavelet Transform and FOAGRNN

为提高心电图分类的准确度,提出一种基于复合特征和FOAGRNN的心电图分类方法.该方法首先用核独立主元分析(KICA)对心电信号进行非线性特征提取得到特征向量A,其次采用小波包变换对心电信号进行多尺度分解,提取小波包节点系数重构后的归一化能量组成特征向量B,A和B组合成复合特征向量C作为心电信号特征,再者利用果蝇算法(FOA)优化广义回归神经网络(GRNN)参数构建出FOAGRNN模型,最后利用优化后的分类模型对心电特征进行识别分类.仿真实验结果表明,FOAGRNN分类方法较其它方法具有很高的分类准确度,分类正确率可达到99.0%.

In order to improve the accuracy of ECG classification,an ECG classification method based on comp-osite characteristics and FOAGRNN is proposed.First of all,this method begins by using the Kernel Independen-t Component Analysis（KICA）to get the characteristic vector A from extracting the nonlinear feature of the EC-G signals,then,wavelet packet transform is used to analyze multi-scale decomposition of ECG signals,ex-tracting the normalized enery which is reconstructed from wavelet packet node coefficient to compose the charact-erristic vector B,acting the compound eigenvector C which is composed by A and B as the ECG characteristic-s.Meanwhile,the fruit fly algorithm（FOA）was used to optimize the generalized regression neural network para-meter to construct the FOAGRNN model.Finally,identifying and classifying the ECG characteristics by the opti-mized classification model.The simulation results show that the classification accuracy of FOAGRNN classificati-on method is so higher compared with other methods,the classification accuracy can reach 99.0%.