Research On Classification Algorithm Of ECG Signal Based On Convolutional Neural Network

At present,cardiovascular disease has become one of the most concerned diseases in the world.The number of people who die from cardiovascular disease every year is more than any other cause.Early prevention and diagnosis can effectively save patients’ lives.Electrocardiogram,as a means of detecting the electrical activity of the heart,is widely used in the detection of cardiovascular diseases.Traditional diagnostic methods are doctor-led and diagnose cardiovascular disease through ECG.Doctors rely on their own professional knowledge and clinical experience to subjectively judge the disease.To ensure the accuracy of the diagnosis,doctors need to have a very high professional level and rich clinical experience.In order to improve the accuracy and robustness of the diagnosis of cardiovascular disease,the technology of automatic analysis of ECG signals based on electrocardiogram has attracted more and more attention.The ECG signal automatic analysis technology is mainly divided into two types,one is the ECG signal recognition algorithm based on traditional machine learning,and the other is the recognition algorithm based on deep learning.Both methods use feature extraction on ECG signals and then use classification algorithms to classify.The former method mainly uses humans to manually select features based on certain prior knowledge.Although the extracted features are more representative,the process of manually selecting features is tedious and often does not make full use of all the features of the ECG signal.The latter method is to learn the features in the ECG signal autonomously through a neural network model,and to fuse the process of feature extraction and classification,avoiding the process of manually selecting features,but the features extracted through the neural network cannot be explained.That is,it is not clear what features the model has learned.The ECG signal is a signal with rich time-frequency characteristics.Neural network models often ignore these characteristics.This paper proposes an algorithm based on the combination of wavelet decomposition and convolutional neural network for the above problems.The ECG signal is mapped to each sub-band by wavelet decomposition,so that the characteristics of the ECG signal in the time-frequency domain are more obvious,and then the features are further extracted by the convolutional neural network for classification.In addition,convolutional neural networks have excellent performance on images and can capture the characteristics of two-dimensional data.Most of the methods that use convolutional networks to process ECG signals are models that use one-dimensional convolution.Contains a dimension of ECG signal expression.In this paper,wavelet decomposition not only makes the time-frequency characteristics of the ECG signal more prominent,but also the data becomes a multi-dimensional expression.Using a two-dimensional convolution model,the characteristics of each sub-band are fully considered to classify the disease’s types.Finally,on the 2017 Cardiology Challenge Contest dataset,experiments were performed on the classification of four types of ECG signals: normal heart rhythm,atrial fibrillation rhythm,other heart rhythms,and noise signals.The total score of F1 for the classification of the four types of signals was 0.76.The model using the one-dimensional convolutional neural network scored about 4 percentage points higher,and the F1 score of the AF rhythm was about 5 percentage points higher.The experimental results prove that the time-frequency characteristics of ECG signals are extracted by wavelet decomposition,and the features are further extracted by using a two-dimensional convolutional network and then classified,which improves the recognition performance of ECG signals.