微量农药溶液近红外光谱PLS模型的不同预处理方法对比研究

比较了常用的多种预处理方法对近红外光谱技术(NIR)检测微量农药溶液含量的影响,使用偏最小二乘法(PLS)分别对各预处理方法处理后的数据建立了数学模型,利用两种不同农药初步探讨了数据量对PLS建模结果的影响。结果表明:矢量归一化(SNV)对26个微量毒死蜱溶液样本的预处理效果综合参数最好,校正集参数为R=0.9957,RMSECV=0.182,预测集参数为R=0.9992,RMSEP=0.0802;减去一条趋势线对20个微量炔螨特溶液样本的预处理效果综合参数最好,校正集参数为R=0.9925,RMSECV=0.649,预测集参数为R=0.9952,RMSEP=0.646。26个样本的微量毒死蜱溶液PLS建模结果优于20个样本的微量炔螨特溶液。     

烟草中淀粉近红外光谱变量的筛选及校正模型的建立

为提高校正模型的预测精度,以烟草中淀粉近红外光谱(NIR)校正模型为研究对象,分别利用全光谱波段(FS)、方差光谱(VS)筛选光谱变量和遗传算法(GA)筛选光谱波长,结合偏最小二乘法建立校正模型(FS+PLS、VS+PLS和GA+PLS),并对100个初烤烟叶样品进行了预测。结果显示:①FS+PLS(变量数1557个)、VS+PLS(变量数781个)和GA+PLS(变量数72个)3种校正模型的决定系数R_c2、交互验证均方根误差(RMSECV)分别为0.9764、0.433,0.9871、0.332和0.9885、0.314。②与FS+PLS和VS+PLS模型相比,GA+PLS模型的光谱变量数分别减少为FS和VS变量数的4.62%和9.22%,主因子数由15降至12,R_c2明显提高,RMSECV显著降低。③FS+PLS、VS+PLS和GA+PLS模型对100个初烤烟叶样品的预测结果显示,R_p2、预测均方根误差(RMSEP)分别为0.9652、0.780,0.9843、0.501和0.9853、0.496,预测值与其对应的化学检测值之间通过配对T检验,显著性Sig.值、T值和平均相对误差(%)分别为0.271、1.107、17.48%,0.973、0.034、13.13%和0.722、0.357、13.12%,3种方法所建立校正模型的预测值与检测值之间均无显著性差异,模型预测精度(RSD)分别为10.34%、6.98%和4.76%。基于逐步优化光谱信息法建立的GA+PLS校正模型的预测精度优于FS+PLS和VS+PLS模型,该方法对于提高复杂化学体系模型的精度有参考意义。      

基于近红外光谱法快速测定许氏平鲉脂肪和 水分含量的方法研究

目的建立近红外光谱法结合偏最小二乘法测定许氏平鲉鱼肉中的脂肪和水分含量,以期简便、快速地对许氏平鲉进行品质分析与评价。方法采用常规分析手段测定70个样品的脂肪和水分含量,同时采集其近红外光谱数据,结合偏最小二乘法(partial least square,PLS)建立许氏平鲉鱼肉中脂肪和水分的定量预测模型,并对比不同光谱预处理方法、光谱范围和因子数对定量预测模型的影响。结果光谱经Savitzky-Golay(S-G)和标准正态变量变换(standardized normal variate,SNV)预处理后,在5341.85~4007.36 cm~(-1)、6556.79~5345.71cm~(-1)和8651.10~7162.33 cm~(-1)光谱范围内,选取主因子数10,建立脂肪的校正模型性能最优;光谱经过SNV预处理后,在8886.38~4061.35cm~(-1)光谱范围内,分别选取主因子数为9时,建立的水分的校正模型性能最优。脂肪和水分含量相对最优PLS模型的校正集相关系数分别为0.9918和0.9912,校正标准偏差分别为0.2680和0.3300,交叉验证相关系数分别为0.9820和0.9810,交叉验证均方差分别为0.3980和0.4850,验证集相关系数分别为0.9804和0.9798,验证集均方差分别为0.3260和0.3070。结论本方法可较为准确地预测许氏平鲉鱼肉中的脂肪和水分含量,能够满足快速分析评价许氏平鲉品质的要求。     

贮藏期内富士和粉红女士苹果品质的FT-NIR无损检测

为探索傅里叶近红外光谱快速无损检测贮藏期苹果品质的方法,在苹果贮藏过程中,每隔30d采集富士和粉红女士(各40个)2个苹果品种共计400个样本的近红外图谱(12000～4000cm-1),用OPUS-QUANT软件预处理光谱,用偏最小二乘法建立通用于2个品种的可滴定酸(TA)、pH值和可溶性固形物(SSC)的数学模型。结果表明：富士和粉红女士的光谱经矢量归一化预处理后,在波段7502～4247cm-1内所建立的可滴定酸模型稳定性较好,该模型校正时的相关系数(R2)和评估均方误分别为0.9231和0.0263%,预测时的相关系数R2和内部交叉验证均方根差分别为0.9071和0.0266%;在波段11995～4247cm-1内,光谱经一阶导数预处理后所建立的pH值预测模型稳定性较好,该模型校正时的R2和评估均方误分别为0.9263和0.0700,预测时的R2和内部交叉验证均方根差分别为0.9113和0.0772;近红外光谱经最大-最小归一化预处理后,在波段6102～5446cm-1所建立的SSC模型效果较好,该模型校正时的R2和评估均方误分别为0.9212和0.3570%,预测时的R2和内部交叉验证均方根差分别为0.9130和0.370%。在富士和粉红女士贮藏期品质检测过程中,建立的通用于这2个品种的TA、pH值和SSC检测的数学模型,稳定性较好,能满足品质快速无损检测的要求。     

Dynamic prediction models for alkaloid content using NIR technology for the study and online analysis of parching in Areca Seed

The goal of this research was to develop dynamic prediction models for contents of arecoline, arecaidine and guvacine by NIR for the study and online analysis of the parching process in Areca Seed (AS). Twenty types of AS were selected from 60 types obtained from various places and were then parched. With partial least squares (PLS), calibration models were generated based on Multiplicative Scatter Calibration (MSC) for guvacine and arecoline and First Derivative + MSC for arecaidine. The root mean square errors of cross-validation (RMSECV) for arecoline, arecaidine and guvacine were 0.141, 0.0822 and 0.181 mg/g, respectively; the root mean square errors of prediction (RMSEP) were 0.224, 0.0897 and 0.187 mg/g, respectively; the correlation coefficients (R) were 0.9813, 0.9658 and 0.9831, respectively. Furthermore, the time–temperature-content-drug efficacy law was analyzed, and some new technology and methods were used in online analysis and quality control.     

间隔连续投影算法应用于近红外光谱苹果糖度模型的优化

目的采用一种改进的连续投影算法(successive projection algorithm,SPA)筛选光谱区间变量,优化苹果近红外光谱模型。方法试验以半透射方式无损地获取134个苹果的光谱信息,再以标准方法破坏性检测其内部糖度指标,在光谱信息与糖度指标之间构建定量模型。区间连续投影算法(intervals SPA, iSPA)是根据各光谱区间之间的投影关系,选择那些具有共线性小的区间变量来构建偏最小二乘模型(partial least square,PLS)。尝试以全区间光谱划分的间隔数量从5到60,步长为5,以优化共线性小的间隔组合。结果当划分为20个间隔时,构建的PLS模型相比于其他划分间隔时的模型,具有较小的交互验证均方根误差和较少的入选变量,此时对预测集的预测均方根误差为0.521,优于常规连续投影算法线性回归和全区间PLS模型的预测性能。结论区间连续投影算法可用于光谱区间变量的筛选,结合偏最小二乘法可提高模型的预测性能。     

红外光谱快速测定油茶籽油中主要功能活性成分含量的模型建立与评价

以从企业采集的50个油茶籽油样品为试验材料,通过扫描获取红外光谱并筛选特征波段,利用偏最小二乘法(PLS)建立油茶籽油中甾醇、维生素E和类胡萝卜素含量的预测模型,并通过系列参数对模型进行评价。结果表明:在400~1 850 cm~(-1)波数范围内,甾醇、维生素E和类胡萝卜素校正集相关系数(R_C)分别为0.978 9、0.980 1和0.949 9,交叉验证均方根误差(RMSECV)分别为42.38、25.64、0.84 mg/kg,经对模型进行验证,上述3种成分预测集相关系数(R_P)分别为0.993 4、0.997 4和0.959 0,预测均方根误差(RMSEP)分别为13.31、6.24、0.18 mg/kg,RPD分别为7.742、12.696和2.889。可见,模型的预测效果较好,说明红外光谱法可应用于油茶籽油中甾醇、维生素E和类胡萝卜素等功能活性成分含量的快速检测。     

Effect of particle size and spectral sub-range within the UV-VIS-NIR range using diffuse reflectance spectra on multivariate models in evaluating the severity of fusariosis in ground wheat

Control (crops grown in natural conditions) and Fusarium head blight (FHB) damaged (crops inoculated with Fusarium culmorum conidia) grain of four wheat cultivars was ground and sieved into three fractions of different particle size. A series of blended samples differing in content of damaged material were prepared within fractions and cultivars, and diffuse reflectance spectra recorded within the 200–2500 nm wavelength range. Partial least-squares (PLS) models for the percentage of damaged material in blended samples were built for each of twelve series within different spectral ranges, and the root-mean-squared error of cross-validation (RMSECV) was used for the assessment of model performance. Errors using the models were lowest for the finest fraction independent of spectral range; however, their values depended on the cultivar. RMSECV for the finest fraction averaged over cultivars ranged from a little below 3.0 (when the ultraviolet light sub-range was used or participated with another one) to 8.1% (when only the near infrared (NIR) sub-range was used). For the medium and coarse fractions, averaged errors showed the same tendency of dependence on the sub-range(s); however, with higher values that increased with an increase in particle size. In conclusion, within the different fractions of particle size and spectral ranges, the most sensitive to the presence of damaged material were models developed for the finest fraction and when the ultraviolet light sub-range was used in modelling.     

可见-近红外高光谱成像技术对灵武长枣VC含量的无损检测方法

为探究基于高光谱成像技术预测灵武长枣VC含量的可行性并寻找最佳预测模型。采集100?个长枣样本在波长400～1?000?nm处的高光谱图像,对光谱数据进行预处理;应用遗传算法（genetic algorithm,GA）、连续投影算法（successive projection algorithm,SPA）和竞争性正自适应加权（competitive adaptive reweighted sampling,CARS）算法对原始光谱数据提取特征波长;分别建立基于全光谱和特征波长的偏最小二乘（partial least squares regression,PLS）和最小二乘支持向量机（least squares support vector machine,LSSVM）VC含量预测模型。结果表明,采用标准正态变换预处理算法效果最优,其PLS模型的交叉验证相关系数为0.839?5,交叉验证均方根误差为16.248?2;利用GA、SPA和CARS从全光谱的125?个波长中分别选取出12、5?个和26?个特征波长;基于CARS建立的PLS模型效果最优,其Rc、Rp、校正均方根误差、预测均方根误差分别为0.896?2、0.889?2、10.746?2%、12.145?3%。研究结果表明基于高光谱成像技术对灵武长枣VC含量的无损检测是可行的。     

PLS法和PCA法在近红外光谱定量分析中的应用研究

利用啤酒的近红外光谱数据比较了 PLS(偏最小二乘法,partial-Squares)和 PCA(主成分回归法,principal componentregression)两种方法在近红外光谱定量分析中的应用。并应用所建模型预测了 21 个啤酒样品麦芽的含量,结论为两种方法均适合近红外光谱定量分析,PLS 法所得预测结果准确度更高。     

应用近红外光谱技术快速测定粳稻品种的直链淀粉含量

应用近红外光谱法以稻谷、糙米、精米、糙米粉和精米粉为扫描材料分别建立了粳稻直链淀粉含量的预测模型。结果表明采用光谱预处理的校正效果比不采用预处理的好,用偏最小二乘法(PLS)获得的粳稻稻谷、糙米、精米、糙米粉、精米粉的回归模型和交叉验证结果为:最优校正决定系数(R2)和交叉检验均方误差(RMSECV)分别为0.8136、2.74,0.8864、2.56,0.8915、2.59,0.9261、2.26,0.9505、1.83,粉碎性样品的误差比整粒样品的误差小。育种实践中,低世代可选用糙米、高世代可选用糙米粉或精米粉作为扫描样本测定稻米直链淀粉含量。     

近红外光谱结合区间偏最小二乘法应用于花生油酸价的测定

采用近红外光谱法结合不同区间偏最小二乘波长筛选法建立花生油酸价的定量分析模型。采用酸碱滴定法测定花生油样本的酸价同时采集近红外光谱数据;采用区间偏最小二乘法(iPLS)、向后区间偏最小二乘法(BiPLS)、移动窗口偏最小二乘法(mwPLS)优选光谱特征区间;采用偏最小二乘法(PLS)对优选出来的谱段建立酸价的定量模型。结果表明,采用mwPLS选择的谱段建立的模型预测效果最佳,RMSECV和RMSEP分别为0.247 76和0.131 5,校正相关系数和预测相关系数分别为0.993 2和0.996 9。因此,近红外光谱结合移动窗口偏最小二乘法可以快速准确测定花生油的酸价。     

近红外光谱法同时测定涪陵榨菜中果胶和总糖的含量

应用傅里叶变换近红外光谱技术,以涪陵榨菜为材料建立与其品质有关的果胶和总糖的定量分析模型。测定50份涪陵榨菜的近红外光谱数据,得到原始光谱,通过光谱预处理方法消除噪声,最后通过偏最小二乘法(PLS)建立回归模型。最终得到涪陵榨菜中果胶和总糖含量的近红外光谱分析模型,其决定系数(R2)分别为98.31、98.35,交叉验证均方差(RMSECV)分别为0.513、0.0531。用该模型对18份未知涪陵榨菜样本进行外部验证,其果胶和总糖的外部验证决定系数(R2)分别为96.69、95.63,预测集标准偏差(RMSEP)分别为0.572、0.0671。内部交叉验证和外部验证均证明,近红外定量分析有较高的准确度,能满足生产中对涪陵榨菜果胶和总糖同时测定的精度要求。     

基于近红外光谱建立纸张表面施胶量测定模型

采集不同施胶量纸张的近红外光谱,利用偏最小二乘法建立测定纸张表面施胶量基于近红外光谱的校正模型。得到校正模型的交叉验证均方差（RMSECV）和外部验证均方差（RMSEP）分别为0.0928和0.1460,校正数据集和独立的检验数据集的预测值与实际测定值之间的相关系数分别为0.9609和0.9294,表明所建立的校正模型具有较高的预测精度和较好的推广性,为纸张无损伤、无预处理的快速、简便、准确的检测提供了新的途径,并且可望实现纸机上的在线检测。     

Prediction of sensory score of Italian traditional balsamic vinegars of Reggio-Emilia by mid-infrared spectroscopy

In this study the potential of Fourier Transform Mid-Infrared (FT-MIR) spectroscopy to predict the sensory score of traditional balsamic vinegar (TBV) of Reggio-Emilia was investigated. The composition of two hundred commercial TBV samples was analysed and the sensory scores, ranging from 133 to 306 points, were evaluated by a certified panel of master experts (reference method). Partial least squares (PLS) regression, obtained from selected pre-processing signal techniques, was used for multivariate calibration to relate the sensory score to the MIR spectra. Performance of different models was compared in terms of coefficient of correlation (r) and root mean square error of cross-validation (RMSECV). The overall best prediction results were obtained using second order derivative with autoscaling and mean-centering of spectral data with the correlation coefficient of 0.889 and 0.885, respectively. It was concluded that the MIR spectroscopy is suitable for rapid instrumental screening of TBV sensory quality.     

基于近红外光谱技术与BP-ANN算法的豆粕品质快速检测

应用近红外漫反射光谱技术结合误差反向传递人工神经网络(BP-ANN)算法,建立豆粕品质(包括水分、粗蛋白、残油)的定量分析模型。将豆粕漫反射吸收光谱数据进行SNV、DT、SG求导、SG平滑和均值中心化处理,然后采用偏最小二乘方法(PLS)降维获取主成分,并优化选择合适的隐含层节点数、隐含层和输出层转化函数,建立校正模型,并用验证样品对校正模型进行验证。结果显示,BP-ANN法建立的水分、粗蛋白和残油的预测相关系数(R)分别为0.981、0.988、0.982,预测标准偏差(SEP)分别为0.120、0.216、0.036,均优于PLS建模方法结果,且满足传统分析方法的重复性要求,表明BP-ANN方法可用于生产过程豆粕品质的快速监控。     

Rapid and non-invasive analysis of deoxynivalenol in durum and common wheat by Fourier-Transform Near Infrared (FT-NIR) spectroscopy

Fourier transform near-infrared spectroscopy (FT-NIR) was used for rapid and non-invasive analysis of deoxynivalenol (DON) in durum and common wheat. The relevance of using ground wheat samples with a homogeneous particle size distribution to minimize measurement variations and avoid DON segregation among particles of different sizes was established. Calibration models for durum wheat, common wheat and durum + common wheat samples, with particle size <500 µm, were obtained by using partial least squares (PLS) regression with an external validation technique. Values of root mean square error of prediction (RMSEP, 306–379 µg kg^–1) were comparable and not too far from values of root mean square error of cross-validation (RMSECV, 470–555 µg kg^–1). Coefficients of determination (r ²) indicated an “approximate to good” level of prediction of the DON content by FT-NIR spectroscopy in the PLS calibration models (r ² = 0.71–0.83), and a “good” discrimination between low and high DON contents in the PLS validation models (r ² = 0.58–0.63). A “limited to good” practical utility of the models was ascertained by range error ratio (RER) values higher than 6. A qualitative model, based on 197 calibration samples, was developed to discriminate between blank and naturally contaminated wheat samples by setting a cut-off at 300 µg kg^–1 DON to separate the two classes. The model correctly classified 69% of the 65 validation samples with most misclassified samples (16 of 20) showing DON contamination levels quite close to the cut-off level. These findings suggest that FT-NIR analysis is suitable for the determination of DON in unprocessed wheat at levels far below the maximum permitted limits set by the European Commission.     

Near-infrared spectroscopy coupled chemometric algorithms for prediction of antioxidant activity of black goji berries (<Emphasis Type="Italic">Lycium ruthenicum</Emphasis> Murr.)

In this study, near-infrared (NIR) spectroscopy coupled with partial least-squares (PLS) regression and various efficient variable selection algorithms, synergy interval-PLS (Si-PLS), backward interval PLS (Bi-PLS) and genetic algorithm-PLS (GA-PLS) were applied comparatively for the prediction of antioxidant activity in black wolfberry (BW). The eight assays were used for quantification of antioxidant content. The developed models were assessed using correlation coefficients (R²) of the calibration (Cal.) and prediction (Pre.); root mean square error of prediction, RMSEP; standard Error of Cross-Validation, RMSECV and residual predictive deviation, RPD. The performance of the built model greatly improved by the application of Si-PLS, Bi-PLS and GA-PLS compared with full spectrum PLS. The R² values determined for calibration and prediction set ranged from 0.8479 to 0.9696 and 0.8401 to 0.9638, respectively. These findings revealed that NIR spectroscopy combined with chemometric algorithms can be used for quantification of antioxidant activity in BW samples.     

A study of beer bitterness loss during the various stages of the Romanian beer production process

Beer is one of the oldest known alcoholic beverages produced by a yeast fermentation of a cereal extract that was germinated in water beforehand. The bitter taste of beer comes from the group of substances introduced during wort boiling, which are the extracted components of hops. The aim of this study was to determine some characteristics of beer (original extract, alcohol content, colour, pH, total acidity, carbon dioxide and bitterness values) during the three stages of the beer production process in a typical Romanian brewery. Measurements were carried out on 60 samples of beers, 10 measurements for each step of the process examining wort, unfiltered fermented beer and bottled beer (final product) from two different types of beer (light and dark). Statistical process control of the beer was performed. Losses in the bitterness units during the production process were between 24.7 and 41.54%, reported in terms of final product. Copyright © 2013 The Institute of Brewing & Distilling