| As the pace and pressure of our life increases, heart disease has become one of the biggest health killers which threat our people’s lives. ECG which can reflect the health status of human heart is applied widely to clinical examination on heart diseases. However, because of the weak electric signals and low anti-interference capacity of the ECG signal,the effect of existing automatic classification algorithms on test set is better than practical clinic performance. It is only as a reference, and the final diagnoses are depended on doctors to finish. So it is essential to study how to improve the accuracy of classification algorithms through which ECG can be detected and identified effectively. In this paper, the main research work lies as follow:Firstly, among the existing methods of classification of ECG, most of the experiments are based on standard database such as MIT-BIH. Because the ECG in standard database comes from a few individuals, which leads to a good performance on the test set and however, a bad effect in the clinical situation. So the CCDD database is used in this paper to improve the generalization ability of the classification algorithms.Secondly, because the normal filtering methods are for single-lead electrocardiogram, through which the useful information is lost. So, aiming at the special structure of multi-lead electrocardiogram and the correlation of each electrocardiogram, the multi-lead electrocardiogram filtering algorithm is used to filter ECG. The loss of useful information is reduced when removing the noise and it is convenient for the follow-up classification and recognition.Thirdly, because of the complexity of ECG signal, the final purpose of classification is to fit a non-linear decision-making function. Features are extracted firstly and then classified in traditional methods. The accurate classification of ECG depends on the accurate features extraction. If the features which are extracted cannot reflect the inner attribute of ECG, the effect of classifying machine will not meet the requirement. So in this paper, features extraction is abandoned and a deep neural network is used to fit a non-linear decision-making function for automatic classifying of ECG based on initial datas.Finally, ECG is classified by the method of CNN. The average classification accuracy of binary classification in normal and abnormal is 82.5%. While classifying the normal ECG and four kinds of arrhythmia diseases ECG, the average classification accuracy is up to 98.82%. The result shows that the accuracy and the generalization ability of the deep neural network classification algorithm is improved in this paper when compared to existing methods. |
PDF Full Text Request