Research On The Method Of Classifying Ⅱ Lead ECG Signals Based On SVM

The diagnosis and classification of arrhythmia is an important research topic.Early diagnosis of arrhythmias is of great significance of reducing the risk and incidence rate of cardiovascular events.At present,there are a lot of research algorithms concerned with the automatic detection and classification of electrical signals,but most of them are usually from the time-frequency domain,morphological features or a direction of statistics to consider the extraction of the characterization features of ECG signals for classification algorithm research,ignoring the effective characterization features in other methods,so it is difficult to get a better and more scientific ECG signal classification effect.To solve the above problems,this paper proposes a method based on Support Vector Machine(SVM)to classify Ⅱ leads ECG signals,including the following research contents:First of all,this paper designs a preprocessing method of the original ECG signal to filter out the interference noise in the ECG signal.In the first step of preprocessing,two continuous median filters are used to remove baseline drift interference from the signal.The second step are to design a low-pass filter to filter the EMG interference contained in the ECG signal.The third step are to design an adaptive filter based on the Least Mean Square algorithm(LMS)to deal with the inevitable power frequency interference in ECG signals.The last step is to divide the whole continuous ECG signal into a single heartbeat.Secondly,in the part of feature extraction of ECG signal,this paper designs different methods to extract the features of ECG signal from three aspects: timefrequency domain,statistics and morphological features.In the time-frequency domain,three methods are designed to extract the time-frequency characteristics of ECG signals,namely,the feature extraction method based on wavelet transforms,the feature extraction method based on LSTM Autoencoder time series and the feature extraction method based on a new Convolutional neural network(CNN)feature extractor.In terms of statistics,a feature extraction method based on skewness and kurtosis is designed.From the morphological point of view,two methods are designed to extract the morphological features of ECG signals,namely,the R-R Interval based feature extraction method and the new morphological feature extraction method.Finally,an integrated classifier based on support vector machine is designed to classify ECG signals using the multi angle representation features extracted above.A single OAO-SVM classification model is constructed for each pair of ECG signal types of an one-against-one(OAO)combination mode.In this paper,different combination decision rules were used to conduct experiments respectively.Following the recommendations of the Association for the Advancement of Medical Instrumentation(AAMI),the heartbeat types of the MIT-BIH arrhythmia database were re divided into normal heartbeat,supraventricular ectopic heartbeat,ventricular ectopic heartbeat and fusion heartbeat,and the training set was divided among patients.Compared with the previous similar integrated classifier research,the integrated classifier designed in this paper has achieved more satisfactory results.The accuracy,precision and recall of the model are 99.05%,96.90% and 91.33% respectively.This method can better complete the task of classification of arrhythmia.