Research Of Low Computational Complexity Algorithm For Heartbeat Classification

Electrocardiosignal(ECG)automatic analysis and diagnosis is currently a hot research topic in the field of signal processing.The technology can effectively improve the efficiency of doctors,promote the development of medical and improve the level of people's health.Heartbeat classification has attracted much attention as an important technology of Electrocardiosignal(ECG)automatic analysis and diagnosis.Although people have achieved a lot of breakthrough in this aspect so far,there are still problems like high computational complexity.Based on the background,this paper researches from two aspects:feature selection and heartbeat multiple classifier.Research on feature selection:combined feature can make accuracy higher,while its dimension is too high and exists feature redundancy because of the relevance.To solve this problem,this paper proposes a heartbeat classification algorithm based on feature fusion by principle component analysis(PCA).Firstly,single feature is normalized.Secondly,different features are combined together to get combined feature.Thirdly,PCA is used on combined feature to obtain lower dimensional fusion feature.Finally,Support Vector Machine(SVM)is used as a classifier to classify different types of heartbeats.By taking time domain feature,Discrete Wavelet Transform(DWT)feature and Discrete Fourier Transform(DFT)as examples,the experiments were performed in MIT-BIH database.The result indicated that heartbeat's classification accuracy was 97.389%when fusion feature's dimension was 100.The accuracy was higher than that of single features,which was 96.5%.It was also higher than that of combined features,which was 97.334%when features'dimension was 492.New fusion feature has higher accuracy than single features and lower dimension than combined features.Research on heartbeat multiple classifier:Traditional heartbeat multiple classification of SVM exists problems of using too many support vector machines what is more,It uses large training samples leading to occupy large memory space when storing and operating.To solve this problem,this paper proposes a heartbeat multiple classification algorithm based on hierarchical support vector machine.In process of building model,firstly,separabilities of different heartbeats are measured by distance.Secondly,two types of heartbeats,which are the hardest to separate,are selected as the input of Support Vector Machine.Thirdly,the two types of heartbeats are combined together as a new type and then we measure the separability between it and the rest kinds of heartbeats.Finally,we repeat the above process until the model is set up.In the process of training,the work is performed from the top down.The number of training set will become smaller and smaller because the leaf node class samples of support vector machine will be removed after each training process.In this paper,experiments were performed using six types of heartbeats in MIT-BIH database.The result indicated that the accuracy of hierarchical support vector machine was 95.8%.The number of SVM it needed was five,which was smaller than the number of the algorithm of one against one and the algorithm of one against all.At the same time,the number of training set would become smaller and smaller.Compared with traditional support vector machine multi classification algorithm,hierarchical support vector machine model can not only have good classification accuracy,but also reduce the number of support vector machine and memory space occupied by storage and computing.