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
I
摘要
目前心血管疾病已经成为人类所面临的主要疾病之一。心电信号作为反映心
脏状况的可视化生理信号,对心血管疾病预防、诊断和救治具有重要的意义。卷
积神经网络具有自动特征提取与分类准确率高融为一体的优势,因此本文研究了
基于改进卷积神经网络的心电信号分类方法,具体研究内容如下:
1.提出了一种基于改进小波阈值和自适应完备集合经验模态分解(Complete E
nsemble Empirical Mode Decomposition with Adaptive Noise,CEEMDAN)的心电
信号去噪方法。为了剔除心电图(Electrocardiography,ECG)信号中的噪声,该算法
首先对 ECG 信号进行 CEEMDAN 分解得到了一组由高频到低频分布的固有模态
分量(Intrinsic Mode Function,IMF)。然后根据相关系数法,对高频 IMF 分量进
行改进阈值的小波去噪。对于低频 IMF 分量,再通过设定固定阈值,对低于该阈
值的 IMF 分量认为是基线漂移信号,将其剔除。然后将去噪后的 IMF 分量和保留
的 IMF 重构,最后获得干净的 ECG 信号。
2.设计了基于改进残差网络的心电信号分类模型,将传统的残差网络中的卷积
层、池化层用 Inception 模块替换来提取不同尺度级别的信息特征,增加网络对尺
度的泛化能力。同时,通过在残差网络中嵌套残差网络方法,充分融合了底层和
高层的信息特征,来进一步缓解梯度问题对模型的干扰,提升了模型对心电信号
的分类准确率。在 MIT-BIH 数据集进行分类测试,最终获得了 95.1%的总体准确
率。
3.设计了基于改进 VGG 网络的心电信号分类模型。首先为了使 VGG 模型更
契合四分类输出,重新设计全连接层结构,其次为了压缩网络中的参数数量,提
高训练效率,将深度可分离卷积引入网络模型。最后由于 VGG 只能够提取到心电
信号的空间特征而忽略了时序特征,针对这一点引进长短期记忆网络(Long
Short-Term Memory,LSTM)来捕捉心电信号空间特征之间存在的时序信息。实验
结果表明改进后的网络在降低参数的同时提升了分类准确性,达到 93.6%的准确率。
关键词:心电信号;小波阈值;自适应完备集合经验模态分解;深度学习;卷积神
经网络
III
Abstract
At present, cardiovascular disease has become one of the main diseases faced by
mankind. As a visual physiological signal reflecting the condition of the heart, the
ECG signal is of great significance to the prevention, diagnosis and treatment of
cardiovascular diseases. Convolutional neural network has the advantage of
integrating automatic feature extraction and high classification accuracy. Therefore,
this paper studies the ECG signal classification method based on improved
convolutional neural network. The specific research content is as follows:
1. An ECG signal denoising method based on improved wavelet threshold and
adaptive complete set empirical mode decomposition (Complete Ensemble Empirical
Mode Decomposition with Adaptive Noise, CEEMDAN) is proposed. In order to
eliminate the noise in the electrocardiography (ECG) signal, this method first
performs CEEMDAN decomposition of the ECG signal to obtain a set of Intrinsic
Mode Function (IMF) from high frequency to low frequency distribution, and then
according to the correlation coefficient method, Perform wavelet denoising with
improved threshold for high-frequency IMF components. For the low-frequency IMF
components, by setting a fixed threshold, the IMF components below the threshold
are considered to be baseline drift signals, and they are removed, and then the
denoised IMF components and the retained IMF are reconstructed to obtain a clean
ECG signal.
2. Designed an ECG signal classification model based on an improved residual
network, and replaced the convolutional layer and pooling layer in the traditional
residual network with the Inception module to extract information features of
different scale levels, and increase the network's generalization ability to scales . At
the same time, by nesting the residual network method in the residual network, the
information characteristics of the bottom and high-level are fully integrated to further
alleviate the interference of the gradient problem on the model, and improve the
accuracy of the model's classification of ECG signals. The classification test was
IV
performed on the MIT-BIH data set, and finally an overall accuracy rate of 95.1% was
obtained.
3. The ECG signal classification model based on the improved VGGNet network is
designed. First, in order to make the VGG model more suitable for the four-category
output, the fully connected layer structure is redesigned. Second, in order to compress
the number of parameters in the network and improve the training efficiency, the deep
separable convolution is introduced into the network model. Finally, because VGG
can only extract the spatial characteristics of the ECG signal and ignore the timing
characteristics, a Long Short-Term Memory (LSTM) is introduced to capture the
timing information between the spatial characteristics of the ECG signal. The
experimental results show that the improved network improves the classification
accuracy while reducing the parameters, reaching an accuracy rate of 93.6%.
Keywords: ECG signal; wavelet threshold; CEEMDAN; deep learning; con
volutional neural network
V
目录
1
绪论 ........................................................................................................................... 1
1.1 研究背景及意义................................................ 1
1.2 心电信号去噪研究现状.......................................... 2
1.3 心电信号分类研究现状.......................................... 3
1.4 本文研究内容.................................................. 4
1.5 论文组织结构.................................................. 5
2
心电信号基本理论 ................................................................................................... 7
2.1 心电信号的产生机理............................................ 7
2.2 心电信号的特征................................................ 8
2.3 心电信号的组成................................................ 8
2.4 心律失常..................................................... 10
2.5 MIT-BIH 心律失常数据库 ...................................... 11
2.6 本章小结..................................................... 13
3
基于改进小波阈值-CEEMDAN 算法的心电信号去噪研究 .............................. 15
3.1 引言......................................................... 15
3.2 ECG 噪声种类 ................................................ 15
3.3 小波变换..................................................... 16
3.3.1 连续小波变换 ............................................................................... 16
3.3.2 离散小波变换 ............................................................................... 17
3.4 CEEMDAN 算法 .............................................. 17
3.4.1 EMD 原理 ..................................................................................... 17
3.4.2 EEMD 原理 ................................................................................... 18
3.4.3 CEEMDAN 原理 ........................................................................... 19
3.5 改进小波阈值-CEEMDAN 算法的实现 ........................... 20
3.5.1 改进小波阈值函数 ........................................................................ 20
3.5.2 小波基函数和阈值的选取 .............................................................. 23
3.5.3 算法实现框图 ............................................................................... 27
3.6 实验结果及分析............................................... 28
3.7 本章小结..................................................... 33
VI
4
基于改进残差网络的心电信号分类识别 ............................................................. 35
4.1 引言......................................................... 35
4.2 人工神经网络................................................. 35
4.3 卷积神经网络的主要特点....................................... 36
4.4 卷积神经网络的基本结构....................................... 37
4.4.1 卷积层 .......................................................................................... 37
4.4.2 池化层 .......................................................................................... 38
4.4.3 全连接层和输出层 ........................................................................ 39
4.4.4 激活函数 ....................................................................................... 39
4.5 改进的残差网络模型........................................... 42
4.5.1 残差网络模型设计 ........................................................................ 42
4.5.2 Inception 模型 ................................................................................ 43
4.5.3 改进残差卷积网络原理 ................................................................. 44
4.6 实验与结果分析............................................... 45
4.6.1 实验数据................................................... 45
4.6.2 评判标准................................................... 47
4.6.3 实验结果分析............................................... 47
4.7 本章小结..................................................... 49
5 基于改进 VGG 网络的心电信号分类识别 .......................................................... 51
5.1 引言......................................................... 51
5.2 深度可分离卷积............................................... 51
5.3 长短期记忆网络 ............................................... 53
5.4 改进的 VGG 网络模型 ......................................... 54
5.4.1 VGG 网络结构 .............................................. 54
5.4.2 VGG 网络模型改进 .......................................... 55
5.5 实验结果及分析............................................... 56
5.6 本章小结..................................................... 58
6 总结和展望 ............................................................................................................. 59
6.1 总结......................................................... 59
6.2 展望......................................................... 60
参考文献 ..................................................................................................................... 61
剩余78页未读,继续阅读
资源评论
2201_75761617
- 粉丝: 20
- 资源: 7339
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