| In recent years,cardiovascular disease has become the disease with the highest fatality rate in the world,and the number of people who die from cardiovascular disease every year is increasing.In clinical medicine,the electrocardiogram(ECG)is a very effective and non-invasive diagnostic tool for cardiovascular diseases.This is because the ECG can record the electrical activity of the heart,so as to correctly reflect the abnormal heart activity.In the past,the interpretation of ECG relied heavily on the experience of clinicians.Obviously,this method is very inefficient,and there is a subjective possibility of misdiagnosis or missed diagnosis.Therefore,it is of great practical significance to develop a classification method that can identify abnormal ECG signals in a timely and accurate manner.This article first proposes an ECG signal preprocessing method.Then based on convolutional neural network(CNN)and Long Short-Term Memory(LSTM)neural network,an improved Siamese neural network algorithm is proposed for the recognition of ECG signals.Finally,two classic ECG signal databases,PTB myocardial infarction database and MIT-BIH arrhythmia database,were used to verify the proposed method.The main research contents of this thesis are as follows:1)Research on the preprocessing of original ECG signals.First,a hybrid noise removal algorithm is proposed to remove the noise interference in the ECG signal.Secondly,a data balance method based on Fourier transform is proposed for the extremely unbalanced data in the MIT-BIH arrhythmia database.Finally,the proposed preprocessing method is verified,and after comparative analysis,it is found that the proposed ECG signal preprocessing method has significant effects.2)Research on classification algorithms for ECG signals.After the original ECG signal is preprocessed,an effective algorithm is needed to classify it.First,the PATS activation function is proposed to address the shortcomings of the traditional activation function.This function is a smooth non-monotonic function with parameters,which reduces the risk of model overfitting to a certain extent.After improving the traditional LSTM network,an ITLSTM network is proposed,which effectively reduces the risk of premature saturation.Finally,this paper constructs the Siamese CNN and Siamese ITLSTM network models for the automatic classification of ECG signals.3)Analysis and research on the results of algorithm experiments.First,a 5-fold cross-validation and control experiment strategy was used to verify the effectiveness of the Siamese CNN model and Siamese ITLSTM network model on the PTB dataset.The results show that for the Siamese CNN model,it achieves an average accuracy of 99.92%,an average sensitivity of 99.92%,and an average specificity of 99.91%.For the Siamese ITLSTM network model,the average accuracy,average sensitivity and average specificity obtained are all over 99%.After that,the 10-fold cross-validation and control experiment strategies were used on the MIT-BIH data set to verify the effectiveness of the two network models,and the models also achieved good classification results. |
