经验模态分解(EMD)方法是一个以信号极值特征尺度为度量的时空滤波过程,它充分保留了信号本身的非线性和非平稳特征,在信号的滤波和去噪中具有较大的优势。文章在介绍EMD分解方法的基础上,结合EMD的多尺度滤波特性,提出了一种新的拉曼光谱去噪方法——EMD阈值去噪法。该方法首先对含噪的拉曼光谱信号做EMD分解,得到各阶本征模态函数(IMF),然后对高频的IMF分量用阈值法进行处理,把经过阈值处理后的高频IMF分量与低频IMF分量叠加得到重构的信号,即去噪信号。通过处理对二甲苯的拉曼光谱信号,分析了在不同噪声水平上不同去噪方法的处理效果。实验结果表明EMD阈值去噪法有效地去除了噪声,较好地保留了光谱的细节信息,与小波阈值去噪方法相比较具有自适应的优势,在拉曼光谱去噪中有很好的应用前景。

The multiresolution empirical mode decomposition (EMD) is a new method of analyzing signal.It can be interpreted as a temporal and spatial filtering based on the extremum characteristic scale of the signal.This method can preserve the nonlinearity and non-stability of signal,and has potential superiority in filtering and denoising.In the present paper,a new denoising method of Raman spectra on the basis of empirical mode decomposition with multiresolution filtering is presented.First,Raman spectra polluted by the white noise is decomposed into several intrinsic mode function (IMF) components of different time scale based on empirical mode decomposition (EMD).Then,the IMF components of high frequencies are preprocessed using the threshold method,and we add these IMF components to the IMF components of low frequencies to achieve denoising signal.For various noise levels,the effects of three methods (the EMD threshold denoising,the wavelet threshold denoising and the EMD low-pass filtering) are analyzed by processing the noisy p-xylene spectra.The results show that the EMD threshold denosing method eliminates the noise effectively.In addition,this method also preserves the detailed information of the original spectra well.By contrast with the wavelet threshold denosing method,the EMD threshold denoising requires no prior knowledge about the sample composition and no selection of the suitable decomposition level,and the adaptation is in evidence. The new method will have a good application prospect in Raman spectra denoising.