小波变换去噪应用于鲜枣糖度近红外光谱检测的研究  

Study on fresh jujube sugar content using near infrared spectroscopy based on wavelet transform denoising

作　　者：马毅[1] 汪西原[1] 雍慧[1] 

机构地区：[1]宁夏大学物理电气信息学院,宁夏银川750021

出　　处：《计算机与应用化学》2011年第3期303-306,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(60961004)

摘　　要：导数光谱可消除光谱背景干扰和基线漂移等因素影响,提高光谱分辨率;但在增强信号时,也使噪声得到增强。本研究依据小波变换消噪的基本原理和方法,在波数为3999.64～11995.06cm^(-1)范围内采集鲜冬枣样品的光谱数据,分别采用Haar、Daubechies、Coiflets和Symlets,4种小波函数在默认阈值的情况下,结合平滑去噪方法,对12个鲜冬枣样品的一阶导数光谱数据进行消噪比较研究和分析。结果显示:采用db4小波函数、分解尺度为3时,去噪效果最好;结合平滑处理方法,单个样品的信噪比(SNR)和均方根误差(RMSE)分别达31.351和0.000011917;12个代表样品的去噪效果具有一致性。研究表明:小波变换能够有效去除导数光谱中的噪声,保留光谱中的有效信息,提高光谱信噪比,不失为1种有效的去噪方法,有助于提高光谱分析精度和后续预测模型的建模准确度。Derivative can correct baseline and background effects, increasing the resolution of the spectra. However, it increases the noise at the same time. According to the basic principles and methods of the wavelet transform signal denoising, the spectra of 12 fresh jujube samples were measured at 3999.64～11995.06 cm^-1 and analyzed in this paper. The first derivative spectra of the samples were denoised by 4 wavelet functions of Haar, Daubechies, Coiflets and Symlets with a default threshold combined with smoothing denoising. The results showed that the noise were eliminated furthest when the wavelet function.is db4-3, and achieved31.351 of signal to noise ratio （SNR） and 0.000011917 of root mean square error （RMSE）, the results were steady across the 12 sample spectra. It can be inferred from the study that the wavelet transform can effectively remove the noise from the derivative spectrum, retain the useful information and improve SNR of spectrum, thus the method can be looked as an ideal method which contribute to improve the accuracy of spectral analysis and modeling accuracy of the prediction model.

关 键 词：近红外光谱 小波变换 鲜枣 粕搜 消噪 

分 类 号：O657.3[理学—分析化学]