基于EMD分解的拉曼光谱小波去噪方法  

Wavelet Denoising Method for Raman Spectroscopy based on EMD Decomposition

作　　者：赵肖宇[1] 贺燕 佟亮[2] 蔡立晶[1] 尚廷义[1] Zhao Xiaoyu;He Yan;Tong Liang;Cai Lijing;Shang Tingyi(Heilongjiang Bayi Agricultural University,Daqing 163319;Qiqihar University)

机构地区：[1]黑龙江八一农垦大学,大庆163319 [2]齐齐哈尔大学

出　　处：《黑龙江八一农垦大学学报》2019年第3期81-86,114,共7页journal of heilongjiang bayi agricultural university

基　　金：黑龙江省科学基金(QC2015071);国家留学基金委(201508230120);中国博士后基金面上项目(2017M620123);黑龙江八一农垦大学博士科研启动基金(XDB-2016-19);黑龙江八一农垦大学科研团队计划项目(TDJH201807);黑龙江博士后资助(LBH-Z18230)

摘　　要：拉曼光谱中尖峰及其临近信号频率极高,常规去噪方法难以区分高频噪声与特征峰信号,所以拉曼光谱去噪一直是该领域内研究热点和难点。针对该问题,提出临界分量判别法,该方法通过计算经验模态分解(empirical mode decomposition,简称EMD)分量的归一化自相关函数,将固有模态分量(intrinsic mode function,简称IMF)划分为噪声主导分量和信号主导分量两部分。根据噪声主导分量和信号主导分量的不同特点,分别使用模极大值方法、软阈值滤波方法处理各分量的小波系数,实现光谱信号去噪。仿真数据去噪实验表明,小波去噪法(1、2阶IMF为噪声主导分量)去噪效果优于其他方法(1 阶IMF为噪声主导分量,1、2、3 阶IMF为噪声主导分量),说明临界分量判别法可以正确识别噪声主导分量和信号主导分量。光谱数据去噪实验表明,应用小波去噪法处理拉曼光谱,信噪比以及均方误差均优于对整条光谱进行模极大值、软阈值和空域相关方法去噪,光谱中噪声几乎得到了完全抑制,突变特征峰信号得到完整保留,获得了最优滤波效果。In Raman spectrum,the characteristic peaks and its adjacent signals’frequency are extremely high.The conventional denoising methods were difficult to distinguish between high frequency noise and characteristic peak signals. Therefore,Raman spectroscopy denoising has been a popular and difficult problem in the field. Aiming at this problem,the critical component discriminant method has been proposed in this paper.The intrinsic mode function(IMF)components from empirical mode decomposition (EMD)were divided into two parts:the noise dominant components and the signal dominant components,based on the normalized selfcorrelation function calculating on the critical component discriminant method.According to the different characteristics of the noise dominant component and the signal dominant component,the modulus maximum value method and the soft threshold filtering method were used to process the wavelet coefficients of each component respectively to denoise spectral signals.The simulation data denoising experiment showed that the method in this paper(the first and second order IMFs are the noise dominant components)was better than other methods (the first order IMF was the noise dominant component,and the first,second and third order IMFs were the noise dominant components).It showed that the critical component discriminant method can correctly identify the noise dominant components and the signal dominant components.Spectral data denoising experiments showed that,the signal-to-noise ratio and the mean square error of Raman spectra denoised by the paper method were better than those of Raman spectra denoised by the modulus maxima,the soft threshold and the spatial correlation method for the whole spectrum.The noises in the spectrum were almost removed,the characteristic peak signals were completely preserved,and obtained the optimal filter indexes.

关 键 词：经验模态分解 小波分析 去噪 拉曼光谱分析 

分 类 号：O657.3[理学—分析化学]