摘要
为更有效对非线性信号进行识别,提出一种经验模态分解神经网络模型,实现经验模态分解算法与卷积神经网络模型的紧耦合。在EMD层利用经验模态分解算法完成信号的自适应分解;引入权重参数,将分解得到的本征模函数依据其对识别的重要性进行自适应加权重构提取特征,增强时域特征提取能力;将提取的特征通过Softmax层完成信号的识别。将该网络模型应用于美国麻省理工学院提供的MIT-BIH心律失常数据库,对心律失常信号的识别准确率为99.38%,高于其它算法的识别准确率,验证了该模型的有效性。
To recognize nonlinear signals more effectively,an empirical mode decomposition neural network model was proposed,which realized the tight coupling between the empirical mode decomposition algorithm and the convolutional neural network mo-del.The empirical mode decomposition algorithm was used in the EMD layer to complete the adaptive decomposition of the signal.With the introduction of weight parameters,the intrinsic mode function obtained by the decomposition was weighted adaptively according to its importance to recognition,so as to realize feature extraction and enhance the ability of time-domain feature extraction.The extracted features were identified through Softmax layer.The network model was applied to MIT-BIH arrhythmia database provided by Massachusetts Institute of Technology.The recognition accuracy of the arrhythmia signal is 99.38%,higher than that of other algorithms,which verifies the validity of the model.
作者
包志强
王美
黄琼丹
吕少卿
BAO Zhi-qiang;WANG Mei;HUANG Qiong-dan;LYU Shao-qing(School of Communication and Information Engineering,Xi’an University of Posts and Telecommunications,Xi’an 710121,China)
出处
《计算机工程与设计》
北大核心
2021年第12期3510-3515,共6页
Computer Engineering and Design
基金
陕西省教育厅科研计划基金项目(17JK0703)
陕西省重点研发计划基金项目(2018GY-150)
西安市科技计划基金项目(201805040YD18CG24-3、GXYD17.5)。
关键词
非线性信号
经验模态分解
卷积神经网络
加权重构
特征提取
nonlinear signal
empirical mode decomposition
convolution neural network
weighted reconstruction
feature extraction