基于神经网络学习算法的胃电信号时频分析  

NEURAL NETWORK BASED ADAPTIVE TIME FREQUENCY ANALYSIS FOR ELECTROGASTROGRAPHY SIGNALS

作　　者：王智顺[1] 李文化[1] 何振亚[1] 杨德治[2] 陈建德 

机构地区：[1]东南大学无线电工程系 [2]南京铁道医学院基础医学部 [3]美国俄克拉荷马城医学研究中心

出　　处：《中国生物医学工程学报》1997年第3期244-252,共9页Chinese Journal of Biomedical Engineering

基　　金：国家攀登计划;美国Witaker生物医学工程基金

摘　　要：本文提出了基于神经网络学习算法的一种新的胃电信号自适应时频分析方法。与传统的短时富里叶变换（ＳＴＦＴ）和Ｗｉｇｎｅｒ分布相比，本方法有如下优点：１、为更好地跟踪胃电信号的变化，提取其时频特征，我们首先引入调制性基函数作为窗口函数。这种窗口函数包含位移参量、尺度参量和中心频率参量，通过调节这些参量可改变窗口函数的中心位置。窗口形状和中心频率大小，因而保证与分析信号保持最佳匹配；２、由于胃电信号的非稳态和非线性，胃电信号时频参量的变化是非线性的。我们应用神经网络这种最佳非线性估计器来学习这些参量，既可获得较高的估计精度，又有较高的自适应性；３、所提出的方法不仅比ＳＴＦＴ和Ｗｉｇｎｅｒ分布有更高的分辨率，而且其时频能量分布图是清晰的，没有交叉项干扰。A novel adaptive time frequency analyzing methodology based on neural network is presented in this paper. Compared with traditional Short Time Fourier Transform (STFT) and Wigner Distribution (WD), the proposed method has the following advantages over them. 1. In order to pursue the variations of the electrogastrography (EGG) signals and extract their time frequency feature, a modulated basis function is used as the window function which includes such parameters as shifting, scaling and center frequency and so on. Adjusting these parameters, one can change the center, shape and frequency etc., of the window function so that it can optimally match the analyzed signals; 2. Since EGG signals are non stationary and nonlinear, the time frequency parameters of them vary non linearly. Because neural network (NN) is considered as a optimal nonlinear estimator, one can use NN to evolve these parameters mentioned above not only to obtain higher precision, but also has better adaptability; 3. The proposed method has higher resolution than STFT and WD , and the picture of time frequency energy distribution is clear, i.e., there is no cross term interference.

关 键 词：胃电图 神经网络 时频分析 子波变换 

分 类 号：R573.04[医药卫生—消化系统] R318.03[医药卫生—内科学]