没有合适的资源?快使用搜索试试~ 我知道了~
资源推荐
资源详情
资源评论
目录
I
目 录
摘 要 ................................................................................................................ I
Abstract ............................................................................................................ II
1 绪论 ............................................................................................................... 1
1.1 研究背景及意义 ..................................................................................................... 1
1.2 国内外研究现状 ..................................................................................................... 2
1.2.1 心电信号预处理研究现状 ........................................................................... 2
1.2.2 心电信号波形检测研究现状 ....................................................................... 3
1.2.3 心电信号特征提取和分类研究现状 ........................................................... 4
1.3 本文研究内容 ......................................................................................................... 6
1.4 论文结构安排 ......................................................................................................... 7
2 心电信号分类相关基础理论 ....................................................................... 8
2.1 心电图信号的基础知识 ......................................................................................... 8
2.2 心电信号预处理 ..................................................................................................... 9
2.2.1 基于小波变换的降噪 ................................................................................. 10
2.2.2 心电信号节拍分割和归一化 ..................................................................... 11
2.3 心律失常类型 ....................................................................................................... 12
2.4 心电数据库介绍 ................................................................................................... 13
2.5 深度学习方法 ....................................................................................................... 14
2.5.1 卷积神经网络 ............................................................................................. 14
2.5.2 长短期记忆网络 ......................................................................................... 18
2.6 本章小结 ............................................................................................................... 19
3 基于 CNN 与 BLSTM 的心律失常分类 ................................................... 20
3.1 数据预处理及表示 ............................................................................................... 20
3.2 基于 CNN 与 BLSTM 的心律失常分类模型 ..................................................... 20
3.2.1 模型整体架构 ............................................................................................. 20
3.2.2 卷积神经网络模块 ..................................................................................... 22
3.2.3 双向长短期记忆网络模块 ......................................................................... 22
目录
II
3.2.4 网络参数设置 ............................................................................................. 23
3.3 实验与结果分析 ................................................................................................... 23
3.3.1 实验数据 ..................................................................................................... 23
3.3.2 交叉验证 ..................................................................................................... 24
3.3.3 评估指标 ..................................................................................................... 24
3.3.4 实验结果及分析 ......................................................................................... 25
3.3.5 与其他方法对比 ......................................................................................... 26
3.3.6 消融实验 ..................................................................................................... 27
3.3.7 有无去噪对比分析 ..................................................................................... 27
3.4 本章小结 ............................................................................................................... 28
4 基于改进的 HRNet-ECA 的心电信号 R 峰检测与心律失常分类 ......... 29
4.1 HRNet 介绍 ............................................................................................................ 29
4.1.1 HRNet 整体架构 .......................................................................................... 29
4.1.2 特 征 融 合 ................................................................................................. 30
4.2 数据预处理 ........................................................................................................... 31
4.3 基于改进的 HRNet-ECA 的心电信号 R 峰检测与心律失常分类 .................... 32
4.3.1 模型整体架构 ............................................................................................. 32
4.3.2 改进的 HRNet 架构 .................................................................................... 32
4.3.3 ECA-Res 单元 .............................................................................................. 33
4.3.4 训练目标与参数设置 ................................................................................. 34
4.4 实验和结果分析 ................................................................................................... 35
4.4.1 实验数据 ..................................................................................................... 35
4.4.2 交叉验证及评估指标 ................................................................................. 36
4.4.3 实验结果与分析 ......................................................................................... 36
4.4.4 卷积核大小的选取 ..................................................................................... 37
4.4.5 有无 ECA 模块的对比分析 ....................................................................... 37
4.4.6 比较 U-Net 和 HRNet 的性能 .................................................................... 38
4.4.7 与其他方法比较 ......................................................................................... 39
4.4.8 可视化分析 ................................................................................................. 39
目录
III
4.5 本章小结 ............................................................................................................... 40
5 总结与展望 ................................................................................................. 41
5.1 总结 ....................................................................................................................... 41
5.2 展望 ....................................................................................................................... 41
参考文献 ......................................................................................................... 43
致谢 ................................................................................................................. 47
作者简介 ......................................................................................................... 48
摘要
I
摘 要
心律失常是心血管疾病中常见的病症之一,诊断心律失常往往需要通过心电图检
查。然而对心电图的目视检查不仅耗时,而且可能导致误诊,影响疾病的预防及治疗。
因此,需要通过自动分析技术来辅助医生进行心律失常的诊断,从而提高诊断效率和
准确性。目前,实现心律失常自动分类的方法主要有基于传统机器学习方法和基于深
度学习方法。传统机器学习方法分类器的性能很大程度上取决于手工提取特征的质量,
模型泛化能力弱。基于深度学习的分类算法可以自动学习特征,能实现比传统机器学
习算法更好的分类性能。因此,根据心电信号的特征,本文进行了基于深度学习的心
电信号 R 峰检测和心律失常分类方法研究:
(1)针对现有方法预处理时间成本高、对噪声敏感问题,提出了一种基于原始
一维心电信号的心律失常自动分类的方法。该方法首先利用卷积神经网络(CNN)学
习和提取心电信号的形态特征,之后通过双向长短期记忆网络(BLSTM)获取特征中
的时间依赖关系,最后借助 softmax 函数完成心律失常的自动分类任务。方法采用
Mish 函数作为激活函数,使得模型在训练中更为稳定。在公开数据库 MIT-BIH 心律
失常数据库上进行五折交叉验证,评估指标达到了 99.11%的平均准确率,表明该模
型可以有效地提取原始心电信号的重要特征。
(2)针对现有的方法大多数是对单个节拍的分类,提出了一种基于改进的高分
辨率网络(HRNet)和有效通道注意力机制(ECA)的心电信号 R 峰检测与心律失常
分类方法,该方法能够对包含多个心电节拍的心电信号片段进行识别分类。方法首先
将原始心电信号分割成 5 秒时间长的心电信号片段,共 1800 个采样点,然后将这些
片段输入到改进的 HRNet 模型中进行自动学习和分类。通过引入有效通道注意力机
制模块,进一步加强了模型的特征提取和特征选择能力。在 MIT-BIH 心律失常数据
库上进行了相关实验,所提方法在心律失常分类任务上获得了 99.86%的平均准确率,
验证了方法的有效性。
关键词:心电信号;深度学习;心律失常分类;R 峰检测
Abstract
II
Abstract
Arrhythmia is one of the common diseases in cardiovascular diseases. The diagnosis of
arrhythmia often needs ECG examination. However, the visual examination of ECG is not
only time-consuming, but also may lead to misdiagnosis and affect the prevention and
treatment of diseases. Therefore, automatic analysis technology is needed to assist doctors
in the diagnosis of arrhythmia, so as to improve the accuracy and efficiency of diagnosis. At
present, the automatic classification methods of arrhythmias mainly include traditional
machine learning methods and deep learning methods. The performance of traditional
machine learning classifier largely depends on the quality of manual feature extraction, and
the generalization ability of model is weak. The classification algorithm based on deep
learning can automatically learn features and achieve better classification performance than
the traditional machine learning algorithm. Therefore, according to the characteristics of
ECG signal, this paper studies the R-peak detection and arrhythmia classification method of
ECG signal based on deep learning:
(1) Aiming at the problems of high preprocessing time cost and noise sensitivity of
existing methods, an automatic classification method of arrhythmia based on original one-
dimensional ECG signal is proposed. This method first uses the convolutional neural
network (CNN) to learn and extract the morphological features of ECG signals, then obtains
the time-dependent relationship in the features through bidirectional long short-term
memory network (BLSTM), and finally completes the automatic classification of
arrhythmias with the help of softmax function. The mish function is used as the activation
function to make the model more stable in training. The five-fold cross validation is carried
out on the open database MIT-BIH arrhythmia database, and the evaluation index reaches an
average accuracy of 99.11%, which shows that the model can effectively extract the
important features of the original ECG signal.
(2) Aiming at the classification of single beat in most of the existing methods, an ECG
R-peak detection and arrhythmia classification method based on modified high-resolution
network (HRNet) and effective channel attention mechanism (ECA) is proposed. This
method can recognize and classify ECG segments containing multiple ECG beats. Methods
first divides the original ECG signal into 5-second ECG segments with a total of 1800
sampling points, and then input these segments into the improved HRNet model for
automatic learning and classification. By introducing the effective channel attention
mechanism module, the ability of feature extraction and selection of the model is further
strengthened. Relevant experiments are carried out on the MIT-BIH arrhythmia database.
The average accuracy of the proposed method in arrhythmia classification task is 99.86%,
which verifies the effectiveness of the proposed model.
Key words: ECG; Deep learning; Arrhythmia classification; R-peak detection
剩余50页未读,继续阅读
资源评论
- winter09012024-02-06资源不错,内容挺好的,有一定的使用价值,值得借鉴,感谢分享。
2301_77550592
- 粉丝: 17
- 资源: 7164
下载权益
C知道特权
VIP文章
课程特权
开通VIP
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功