一种深度小波过程神经网络及在时变信号分类中的应用

A Deep Wavelet Process Neural Network and Its Application in Time-varying Signal Classification

针对多通道非线性时变信号分类问题,提出一种基于稀疏自编码器的深度小波过程神经网络(SAE-DWPNN)。通过构建一种多输入/多输出的小波过程神经网络(WPNN),实现对时变信号的多尺度分解和对过程分布特征的初步提取;通过在WPNN隐层之后叠加一个SAE深度网络,对所提取的信号特征进行高层次的综合和表示,并基于softmax分类器实现对时变信号的分类。SAE-DWPNN将现有过程神经网络扩展为深度结构,同时将深度SAE网络在信息处理机制上扩展到时间域,扩展了两类模型的信息处理能力。该网络可提取多通道时序信号的分布特征及其结构特征,并保持样本特征的多样性,提高了对信号时频特性和结构特征的分析能力。文中分析了SAE-DWPNN的性质,给出了综合训练算法。以基于12导联ECG信号的7种心血管疾病分类诊断为例,实验结果验证了模型和算法的有效性。

Aiming at the problem of multi-channel nonlinear time-varying signal classification,a deep wavelet process neural network based on sparse self-encoder(SAE-DWPNN)is proposed.By constructing a multi-input/multi-output wavelet process neural network(WPNN),multi-scale decomposition of time-varying signals and preliminary extraction of process distribution features are realized;By superimposing a SAE depth network after the WPNN hidden layer,the extracted signal features are synthesized and represented at a high level,and the time-varying signals are classified based on the softmax classifier.The SAE-DWPNN extends the existing process neural network into a deep structure,and expands the deep SAE network into the time domain in the information processing mechanism.It improves the information processing capabilities of the two models.The network can extract the distribution characteristics and structural features of multi-channel time series signals,and maintain the diversity of sample features,which improve the analysis ability of signal time-frequency characteristics and structural features.In this paper the properties of SAE-DWPNN is analyzed and a comprehensive training algorithm is given.Taking the classification of seven cardiovascular diseases based on 12-lead ECG signals as an example,the experimental results verify the validity of the model and algorithm.