Research On Classification Method Of Electrocardiogram With Wavelet Transform And Neural Network

Following with the rhythm speeding up of urban residents living and working, and dietdisorder, heart disease has become a kind of high incidence disease in modern society,electrocardiogram (ECG) can reflect the work condition of the human heart, it is a veryimportant basis to diagnosis the heart disease. So, it takes more and more researchers’attention how to get not only accurate but also rapid computer automatic classification of thevast amounts ECG data record.This thesis studies the method of ECG classification, which based on the wavelettransform and probability neural network. Because the ECG signals are faint, just mill voltlevel (mV), so it is easily influenced by environment and the noise. If the waveformcharacteristics are directly extracted from the signal before it is filtered, the addition of noisewill affect the quality of the eigenvalues, and will reduce the accuracy and generalizationability of network. How to filter the high frequency interference, which is effected by muscletremors and power frequency interference, and low frequency interference, which is effectedby baseline drift, is very important for the following feature extraction and ECG classification.So this paper arranges like this: firstly, adopt the wavelet algorithm to denoise the frequencynoisefrom the ECG signals, and use mathematical morphology to detect the QRS wave and toextract the twelve basic characteristic value of the ECG, then use the parallel probabilityneural network classification algorithm to do real-time and accurate classification for thediagnosis of abnormal heart rate. Simulation experiment shows that this ECG classificationmethod can achieve above95%accuracy and need less than10seconds. This study of theproposed algorithm improves the speed and accuracy of ECG classification. It has greatsignificance and practical value for the medical personnel to diagnose heart disease quicklyand accurately.