期刊文献+

基于CNN-BiLSTM-Attention神经网络的心电信号分类研究

Research on ECG Signal Classification Based on CNN-BiLSTM-Attention Neural Network
下载PDF
导出
摘要 心电信号分析在心血管疾病的诊断中一直起到至关重要的作用,为了能在不同类型的心电信号中实现自动分类、识别异常心率,研究并提出了一种基于深度学习的分类模型,用来自动识别5种不同类型的心拍。研究首先利用卷积神经网络中的局部感知野特性来提取信号中的局部特征,再结合双向长短期记忆网络捕获心电序列中的前后依赖关系,最后引入注意力机制为提取到的每一个特征赋予区分化的权重,让模型在训练的过程中充分关注被分配了更高权重值的主要特征,增强模型的分类能力。针对类别数据不平衡的问题,利用合成少数过采样技术(SMOTE)进行数据增强,进一步优化了模型的分类效果。研究使用MIT-BIH作为实验数据,通过对实验结果的对比分析,验证了模型在心电信号分类方面的可行性。 ECG signal analysis has always played an important role in the diagnosis of cardiovascular diseases.In order to achieve automatic recognition of arrhythmias in different types of ECG signals,a new classification algorithm model based on deep learning to automatically recognize five different types of heart beats is studied and proposed.The convolutional neural network(CNN)is used to extract local characteristics by the local perception field of neural network,and the bidirectional long-short term memory(BiLSTM)is used to capture contextual dependencies in ecg sequences,and the weight of region differentiation is given to each characteristic signal extracted from model by attention mechanism,which can promote the ability of classification for ECG by focusing on the weights of important features.In terms of the problem of imbalance,synthetic minority oversampling technique(SMOTE)is used to promote the quality of data,which further optimizes the effcet of the model.In this study,the effectiveness of the algorithm is evaluated with the MIT-BIH arrhythmia dataset,and compared with the existing research results the advantage of our model is proved.
作者 袁成成 刘自结 王常青 杨飞 YUAN Chengcheng;LIU Zijie;WANG Changqing;YANG Fei(School of Biomedical Engineering,Anhui Medical University,Hefei 230032;Institute of Plasma Physics,Chinese Academy of Sciences,Hefei 230031;University of Science and Technology of China,Hefei 230026)
出处 《计算机与数字工程》 2022年第11期2478-2484,共7页 Computer & Digital Engineering
基金 国家自然科学基金青年项目(编号:62001005) 安徽省自然科学基金面上项目(编号:2108085MH303)资助。
关键词 心电信号分析 特征提取 卷积神经网络 注意力机制 ECG signal analysis feature extraction CNN attention mechanism
  • 相关文献

参考文献5

二级参考文献18

共引文献3390

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部