Application Of Deep Convolutional Neural Network On ECG Signal Recognition

Arrhythmia is one of the most dangerous cardiovascular diseases caused by abnormal heart rhythm disorders.The electrocardiogram shows changes in the electrical activity of the heart and is an important tool for diagnosing arrhythmia.At present,the recognition of ECG signals mainly relies on ECG experts to identify and diagnose ECG based on their own experience.The complexity and diversity of ECG signal forms,and the time consuming of ECG data acquisition easily lead to low recognition accuracy and poor adaptability.It’s time-consuming,laborious and costly to rely on human to participate in diagnosing.To solve the above problems,in this paper,deep convolutional neural network is applied to the feature extraction and recognition of ECG signals,and in-depth research and detailed analysis on it have been conducted.The main contents include:Illustrate the research status of ECG signal recognition at home and abroad,and analyze the characteristics of ECG signal.The recognition of ECG signal includes data preprocessing,feature extraction,feature selection and classification.This paper has mainly expounded the algorithm of feature extraction and classification on ECG signal recognition,compared and analyzed advantages and disadvantages of deep learning technology respectively,and emphasized the applicability of deep convolutional neural network on the processing of ECG signal.The generation mechanism and waveform characteristics of ECG signals were studied,and the forms of several kinds of ECG signal were analyzed,which laid the foundation for the classification of signals.Preprocess ECG signal data.According to the characteristics of persistent noises and burst noises in ECG signal,responding denoising algorithms were designed to remove noises;the position of R peak was obtained from annotation file of MIT-BIH arrhythmia database,which were used as a baseline to capture the heartbeat;applying normalization algorithm to normalize heartbeat data,unifying data specification and building ECG signal data set.The preprocessing of ECG signals mainly includes signal denoising,signal data segmentation and data normalization.Design the infrastructure of deep convolutional neural network.Considering the strong temporal characteristic of ECG signal,long short-term memory and fully connected layer are applied to be the output of convolutional neural network,through the forget gate and input gate of which features are weaken and strengthen selectively in different time series to capture regularity within the data.To improve performance of convolutional neural network on feature extraction,the attention mechanism was embedded after the pooling layer,which transforms the feature map outputted into attention map through both channel attention module and spatial attention module.The attention map applies different weights to the feature map and conducts another extraction of ECG signal features to refine features adaptively,and features are enhanced from two dimension;the features extracted will be divided into series to conduct LSTM network to focus on the most discriminative part of signal,and the network can completes the expression of the internal relationship between data better.The experimental results shows that the deep convolutional neural network proposed in this paper can extract features of ECG signals effectively,and the classification accuracy reaches 99.23%.The proposed algorithm has a good reference value in clinal application.