APP下载

基于深度堆栈网络的心电信号识别

2021-09-06张锐王茹黄俊曾鑫

哈尔滨理工大学学报 2021年3期
关键词:深度特征

张锐 王茹 黄俊 曾鑫

摘 要:传统的心电信号识别算法依靠心电专家参与特征识别,费时费力,诊断成本高,心电信号形态复杂多样导致识别准确率低、适应性差。为解决上述问题,将栈式稀疏自编码器(SSAE,Stacked Sparse Autoencoder ),与Softmax分类器相结合形成深度堆栈网络(DSN, Deep Stacked Network)完成对心电信号的自动识别。通过3个稀疏自编码器堆叠的方式完成心电信号特征提取,逐层刻画心电信号的高维特征,由Softmax分类器完成心电信号识别。详细评估了深度堆栈网络的模型特性,确定了该网络模型的超参数,训练集样本和测试集样本源于MIT-BIH数据库。实验结果表明采用本文所提方法对心电信号进行识别,总识别率达到99.69%,验证了所提方法的有效性。

关键词:

栈式稀疏自编码器;特征提取;心电信号识别;稀疏参数

DOI:10.15938/j.jhust.2021.03.016

中图分类号: TP183

文献标志码: A

文章編号: 1007-2683(2021)03-0108-07

ECG Signal Recognition Based on Deep Stacked Network

ZHANG Riu1, WANG Ru1, HUANG Jun2, ZENG Xin1

(1.School of Automation,Harbin University of Science and Technology,Harbin 150080,China; 2.Chengdu East Road Traffic Technology Co., Ltd, Chengdu 610037, China)

Abstract:The traditional electrocardiogram (ECG) signal recognition algorithms rely on ECG experts to participate in feature recognition, which is time-consuming and laborious with high diagnostic cost. Complex and diverse ECG signal patterns result in low recognition accuracy and poor adaptability. To solve the above problems, the stack Sparse Autoencoder was combined with the Softmax classifier to form a Deep stack Network to realize automatic recognition of ECG signals. The feature extraction of ECG signals was completed by stacking three sparse autoencoders, and the high-dimensional features of ECG signals were depicted layer by layer, and the ECG signals were identified by Softmax classifier. Detailed assessment of the model characteristic of Deep stacked Network, determine the super parameter of the network model, sample training set and test set samples from MIT/BIH database. The experimental results show that the total recognition rate of the proposed method is 99.69%, which verifies the effectiveness of the proposed method.

Keywords:stacked sparse auto-encoder; feature extraction;  ECG signal recognition; sparse parameter

0 引 言

心血管疾病仍然是全球死亡的主要原因。心电图能提供心脏活动的信息,对各种心律失常的分析诊断具有极为重要的意义。心电信息是患者重要的临床资料,临床诊断需要医生具备高度的信息综合处理能力。但是心电特征具有一过性,所以临床上多用心电图仪对患者进行24 h甚至48 h的连续监控。传统的人工诊断面对庞大的数据困难极大,针对心电信号自动分类的智能算法变得尤为重要。

心电信号的自动分类过程通常包括心电信号采集、心电信号的预处理、心电信号的特征提取和分类。临床采集的心电信号是非常珍贵的,所以现在大多数心电信号的研究者研究所用的心电信号来源于4个国际权威的心电数据库,本次实验采用的是其中之一的MIT-BIH心律失常数据库。……

登录APP查看全文

猜你喜欢

深度特征
如何表达“特征”
不忠诚的四个特征
深度观察
深度观察
深度观察
深度观察
提升深度报道量与质
线性代数的应用特征