形变监测数据的多尺度滤波

Multi-scale Filtering Of Deformation Monitoring Data

尺度是许多物理现象的内在特性,信号含有不同尺度的物理结构特性,随着信号分析的多尺度的变化引起的相应的信号特征的产生,消失,融合,这就是信号对多分辨率的感知,因而通过对信号的多尺度的分解可以很好的检测信号的特征。通过多尺度分解后建立各细节分量的变形分析模型为:H(ti)=f(ti)+β(ti)+εi,其中f(ti)为趋势部分,利用小波分析进行提取,我们可认为小波分析后低频部分为趋势部分信号,β(ti)为隐含周期性或者异方差性部分,平稳或非平稳随机分量,它包含在高频信号部分。利用正弦周期函数和RBF网络建立β(ti)逼近模型。εi表示随机误差部分。

Scale is the inherent property of many physical phenomena,Signals contain physical structure characteristics in different scales,the corresponding signal characteristics producing,disappearing,fusioning with the analysis of the multiscale signal,This is the signal apperceive to the multiresolution,Thus can detect signal characteristics very well based on multi-scale decomposition of the signal.We establish the deformation analysis model of the details through the multi-scale decomposition：H（ti）=f（ti）＋β（ti）＋εi,f（ti）is partly the trend,extracting with the wavelet analysis,We consider the low frequency part of the wavelet analysis is the trend partly signal,β（ti）is steady or non-stationary random component that implicit cyclical or heteroscedastic,which contained in the high frequency part of signal,we establish approximation β（ti） model with sinusoidal periodic function and RBF network,εi express random errors.