Study Of ECG Recognition Based On General CNN And Heartbeat Model

Arrhythmia is a cardiac abnormal beat caused by abnormal electrical activity of the heart,which may threaten the life of a patient.Electrocardiogram(ECG)heartbeat classification has important value and significance for the clinical diagnosis of arrhythmia.However,the performance of the heartbeat recognition model for the recognition and classification of ECG beat is still not ideal.Although the existing ECG heartbeat features generated by deep learning are superior to the heartbeat features generated by traditional methods,there are serious data imbalance problems among the various types of ECG,resulting in the poor performance of the existing heartbeat features generated by the deep learning method.For the problem of uneven data distribution among various types,this paper assumes that each type of ECG in arrhythmia consists of a unified background transformation space.It consists of a projection function that transforms the characteristics of the heartbeat into a general characteristic projection space and personality feature projection space of the heartbeat.Then,this paper proposes a feature generation model based on a general background convolutional neural network(CNN).Establishing high-dimensional spatial CNN model to extract heartbeat features.The high-dimensional spatial features of the heartbeat are divided into two steps.Firstly,the general CNN model that can effectively express the heartbeat common information is constructed on the basis of all kinds of heartbeat isometric data;Then,based on the general CNN model,the category CNN model which can effectively reflect the propensity information of the corresponding heartbeat is constructed on the various heartbeat data.Finally,the heartbeats are identified and classified by integrating the CNN models of each category.The experimental results on the MIT-BIH database show that the average sensitivity of the method is 99.68 %,the average positive detection rate is 98.58%,the comprehensive indicators is 99.12%,and all the indicators are significantly better than the existing methods.Although the performance of the category CNN model method is obviously better than the existing method.The category CNN model method is independently trained for each class of CNN model,so there is a problem that the common information is inconsistent and the class spacing of the category CNN is small.These problems lead to category CNN model is still not perfect and the recognition performance needs further improve.In view of the above problems,based on the above-mentioned category CNN model,this paper proposes a kind of heartbeat feature with higher discrimination and robustness based on minimization of intra class distance and maximization of inter class distance.The specific methods are as follows.Firstly,a general CNN model that can effectively express the common information is constructed on the basis of all kinds of heartbeat isometric data;Then to minimizing the intra class distance and maximizing the inter class distance as the objective function,with a mixture of all kinds of isometric data as the training set,based on the general CNN model,a joint CNN model that can effectively separate the propensity information of all kinds of heartbeats is constructed;Finally,the cross entropy value based on output is integrated to identify and classify.The experimental results on the MIT-BIH database show that all kinds of heartbeat recognize indicators are 100%,which perfectly solves the problem of automatic recognition and classification of ECG heartbeats.The performance of this method is not only better than the existing method,but also better than the class CNN model of this paper.The experimental results on the MIT-BIH database show that all kinds of heartbeat recognize indicators are 100%,which perfectly solves the problem of automatic recognition and classification of ECG heartbeats.The performance of this method is not only better than the existing method,but also better than the category CNN model of this paper.