| As IoT devices are widely used in the collection of ECG data,how to effectively classify arrhythmia automatically,so that patients can be quickly diagnosed and treated,and how to recover clean ECG signals from noisy ECG signals with high quality has become an important research topic in the field of intelli gent medical treatment.The increasing maturity of deep learning provides a strong support for solving this problem.However,in the application of arrhythmia classification based on deep learning models,the existing classification methods do not fully co nsider the influence of the information before and after each time point in the ECG sequence data on the hidden features,and the noise in the ECG signal affects the classification of ECG signal.However,the existing noise reduction methods simplify the types of noise,even if multiple noise types are considered at the same time,the local correlation and global correlation of the ECG signal are not fully considered.In order to solve the above problems,this thesis focuses on the detection and classification of arrhythmia and the intelligent noise reduction method of ECG by using the deep learning technology and methods.Among them,the classification model has achieved a higher accuracy,and the noise reduction method has achieved a lower root mean square error and a higher signal-to-noise ratio.The main contents are introduced as follows :First,a new classification model of arrhythmia detection based on CNN and BiLSTM is proposed(CNN-BiLSTM).In the data pre-processing part,first use preprocessing and adaptive threshold method to complete the ECG signal denoising and QRS wave positioning,and then selecte the heartbeat extraction algorithm to extract different types of heartbeat with the same length.In the part of classification model,CNN model is good at extracting local parallel features and local features of ECG signal.Based on the advantages of LSTM,which is good at extracting the features of long-time series data,the BiLSTM model is used to extract long-time dependent features of ECG signal.The CNN-BiLSTM classification model was validated by selecting MIT-BIH arrhythmia data set.It realizes the automatic detection and classification of four types of arrhythmia: normal sinus rhythm(N),left bundle branch block(L),right bundle branch block(R)and ventricular premature beat(V).Finally,the accuracy of 99.18%,accuracy of 99.18%,recall of 99.18% and F1 score of 99.18% are achieved.The experimental results show that the classification model proposed in this thesis can well predict the time-dependent ECG data.Secondly,a noise reduction method for ECG signal based on Wasserstein distance generation adversarial network(WGAN)is proposed.In order to reflect the real situation of unknown noise types in real life,in the data preprocessing part,three different kinds of noise with SNR of 0d B,1.25 d B and 5d B are superposed on the clean ECG signal,and the noise data and the clean ECG data are synthesized into noisy ECG data.In the noise reduction part of the model,seven types of noise mixture are considered,and from the point of view of local and global characteristics of ECG signal,two additional loss functions are added to the generated network to guide the generation of noise-reducing ECG data.The MIT-BIH arrhythmia standard database and MIT-BIH noise pressure ECG database are uesd to verify the noise reduction model.no matter for single noise reduction or mixed noise reduction,this noise reduction method has achieved low RMSE and SNR.The experimental results show that the intelligent noise reduction model proposed in this thesis can well realize the noise reduction of ECG signal. |
