基于近红外光谱分析技术测定库尔勒香梨硬度

采用近红外(NIR)漫反射光谱法对新疆特色梨果库尔勒香梨的五种不同果（包括青头、粗皮、脱萼、宿萼、突顶果）的硬度进行测定。由于近红外光谱数据量大且原始光谱噪声明显、测定水果时散射严重等导致光谱建模时关键波长变量提取困难。以新疆库尔勒香梨为研究对象,为了有效地消除固体表面散射以及光程变化对NIR漫反射光谱的影响,首先采用标准正态变量变换（SNV）和多元散射校正（MSC）对库尔勒香梨的原始光谱进行预处理。为寻找适合近红外光谱检测库尔勒香梨硬度的最佳特征波长筛选方法,进行香梨近红外光谱的特征波长变量选择方法的比较与研究。研究比较了两种特征波长筛选方法对库尔勒香梨硬度偏最小二乘法（PLS）建模精度的影响。同时使用反向偏最小二乘（BiPLS）和遗传算法结合反向偏最小二乘（BiPLS-GA）在全光谱范围内筛选香梨硬度的特征波长变量,将校正均方根误差（RESMC）、预测均方根误差（RESMP）以及决定系数（R2）作为模型的评价标准,并最终确定最优波段选择方法及最佳预测模型。基于选择的特征波长变量建立的PLS模型（BiPLS-GA）与全光谱变量建立的PLS模型进行比较发现BiPLS-GA模型仅仅使用原始变量中6.6%的信息就获得了比全变量PLS模型更好的库尔勒香梨硬度的预测结果,其中R2,RMSEC和RMSEP分别为0.91,1.03和1.01。进一步与基于反向偏最小二乘算法（BiPLS）获得的特征变量建立的PLS模型比较发现,BiPLS-GA不仅可以去除原始光谱数据中的无信息变量,同时也能够对共线性的变量进行压缩去除,使得建模变量从301个减少到20个。极大地简化模型的同时有效地提高了模型的预测精准度和稳定性。因此该方法能够有效地用于近红外光谱数据变量的选择。证明了近红外光谱分析技术结合BiPLS-GA模型能够高效地选择出建模变量,去除与库尔勒香梨硬度无关的近红外光谱信息,显著地提高库尔勒香梨硬度定量模型的预测精度。这不仅为新疆地区特色梨果库尔勒香梨的快速、精确、无损优选分级提供一定的技术支持,同时也为基于近红外光谱分析技术预测水果内部品质的研究提供了参考。     

近红外光谱法分析油页岩含油率中波长选择方法的研究

波长选择是光谱建模分析的重要步骤。研究了近红外光谱法分析油页岩含油率过程中的波长选择方法,用以剔除光谱数据中的冗余信息和干扰信息,提高分析模型的建模效率和预测能力。分别采用相关系数法(CC)、移动窗口偏最小二乘法(MWPLS)和无信息变量消除法（UVE）对油页岩近红外漫反射光谱数据的波长区间进行了选择,研究了不同阈值、窗口宽度和噪声矩阵对上述方法的影响,建立了所选择波长处的反射率数据和样品含油率标准值间的偏最小二乘（PLS）分析模型,比较了上述方法的选择效果。结果表明：与使用全谱数据建模相比,采用上述方法筛选过的光谱数据均能提高模型的建模效率和预测能力,其中经UVE法筛选后的光谱数据仅占全谱数据总数的22.8%,模型的RMSECV却降低了9.3%,RMSEP降低了4.5%。     

基于离散萤火虫算法的近红外波长优选方法研究

近红外光谱数据量大,需要进行压缩,以降低建立光谱校正模型的计算复杂度,提高模型精度和稳健性。为此,提出了一种基于离散萤火虫算法(discrete firefly algorithm)的近红外光谱波长变量筛选方法。首先采用蒙特卡罗方法剔除异常值,并应用Kennard-Stone法进行校正样本的选择。对通用萤火虫算法进行离散化处理,改进了吸引度的自适应公式,在移动公式中增加了牵引权重,以适应离散化处理的影响和优化算法,并在离散萤火虫算法中加入精英保留策略,加快算法的收敛速度。实验中找到DFA算法中的各项参数中的最佳值。通过离散萤火虫算法优选波长变量,建立发酵液中丁二酸含量的近红外光谱偏最小二乘回归(partial least squares regression)校正模型。与标准遗传算法(genetic algorithm)优选波长方法进行了比较。结果显示,基于离散萤火虫算法的波长优选方法所建立的PLS校正模型,其校正集的相关系数(R2_c)为0.986,RMSEC为0.409,预测集的相关系数(R2_p)为0.969,RMSEP为0.458,模型稳健性和精度都要优于全光谱建模以及遗传算法波长优选方法。显示了DFA在近红外光谱数据筛选方面的优越性。     

Nondestructive Quantification of Foliar Chlorophyll in an Apple Orchard by Visible/Near-Infrared Reflectance Spectroscopy and Partial Least Squares

Chlorophylls respond rapidly to the current physiological status of a tree and reflect nutrient availability. Visible/near-infrared spectroscopy was attempted to determine foliar chlorophyll content in an apple orchard. Backward interval partial least squares and genetic algorithms were sequentially applied to select an optimized spectral interval and an optimized combination of spectral regions selected from informative regions in model calibration. Backward interval partial least squares was used to remove the noninformative regions, which significantly reduced the number of variables. The subsequent application of genetic algorithms-partial least squares to this reduced domain could lead to an efficient and refined model. The performance of the final model was back-evaluated according to root mean square error of calibration (RMSEC) and the correlation coefficient (R _c ) in the calibration set, and was then tested by root mean square error of prediction (RMSEP) and the correlation coefficient (R _p ) in the prediction set. The optimal backward interval partial least squares-genetic algorithms model was obtained with 5 partial least squares factors with 3 spectral regions and 71 variables selected. The measurement results of the final model were achieved as follows: RMSEC = 0.26, R _c  = 0.91 in the calibration set; and RMSEP = 0.22, R _p  = 0.91 in the prediction set. This experiment showed that visible/near-infrared spectroscopy and backward interval partial least squares-genetic algorithms are useful tools for nondestructively assessing foliar chlorophyll content and may have potential application for field assessments in decision-making and operational fertilizer management programs for apple orchards.     

基于近红外特征光谱的番茄苗氮含量快速测定方法研究

为了提高近红外光谱技术快速测定番茄苗氮含量的准确度和稳健性,比较分析竞争自适应重加权采样法(CARS)、蒙特卡罗无信息变量消除法(MCUVE)、向后间隔偏最小二乘法(BiPLS )和组合间隔偏最小二乘法(SiPLS)四种特征波长挑选方法,筛选与番茄苗氮含量相关的特征光谱。在十种不同氮素处理水平下(尿素溶液浓度0～120 mg·L-1),培育60株番茄苗样本(每个处理6株),使其分别处于不同程度的过量氮素、氮素适度、缺氮素和无氮素状态。分别采集每株番茄苗样本的叶片,扫描其12 500～3 600 cm-1波段的近红外光谱。比较四种方法所建立的番茄苗氮素定量分析模型可知：CARS和MCUVE挑选的特征变量所建定标模型的性能比BiPLS和SiPLS挑选的特征变量所建定标模型的性能更优,但是预测性能远低于后者。其中,基于BiPLS建立的番茄苗氮素含量预测模型性能最佳,相关系数(r)、预测均方根误差(RMSEP)和性能对标准差之比(RDP)分别为0.952 7,0.118 3和3.291 0。因此,近红外光谱技术结合特征谱区筛选可以有效地提高番茄苗叶片氮素含量的定量分析模型指标,使模型更实用化。但是,特征波长挑选方法不具有普适性。基于单个波长变量筛选的方法所建立的模型较为敏感,更适用于样本状态较为均匀的待测对象;而基于波长区间筛选的方法所建的模型相对抗干扰性更强,更适用于样品状态不均匀,重现性较差的待测对象。因此,特征光谱筛选只有与样本状态及建模指标结合,才能使其在建模过程中发挥更好的作用。     

连续投影算法及其在小麦近红外光谱波长选择中的应用

采用全谱建立多元校正模型时,通常计算量大,模型不够稳健,而且模型的预测精度往往也不能达到最优。文章介绍一种新的波长选择方法：采用连续投影算法(successive projections algorithm),并将其集成偏最小二乘(partial least squares)多变量校正技术构成SPA-PLS方法,用于谷物小麦近红外光谱波长优化选择及其与水分含量的定量分析。结果表明：在经SPA算法后,光谱波数可削减97.72%,后继的定量校正模型结构得到显著简化,模型的稳健性也大大增强;同时,被选取的波长物理意义明确,模型的解释能力增强,而模型的预测性能也与GA-PLS方法相当。     

净信号预处理结合径向基偏最小二乘回归在血糖无创检测中的应用

为了提高人体血糖近红外光谱定量分析模型的预测精度,结合净信号预处理(NAP)算法和径向基偏最小二乘(RBFPLS)回归建立了一种适合于人体血糖测量的非线性建模方法NAP-RBFPLS。本文首先利用NAP对近红外光谱进行预处理来有效地提取原始光谱中仅与葡萄糖信号相关的光谱信息,从而有效地减弱了人体血液中水、白蛋白、血红蛋白、脂肪等成分的吸收干扰以及人体体温的变化、测量仪器本身的漂移、测量环境的变化和测量条件的变化引起的干扰因素与血糖变化的偶然相关问题;然后把净信号预处理后的近红外光谱数据通过RBFPLS建立了非线性定量分析模型来解决由于人体强散射引起的血糖浓度与近红外光谱之间的非线性关系,并与偏最小二乘(PLS)、基于净信号预处理的偏最小二乘(NAP-PLS)和RBFPLS这三种建模方法建立的定量分析模型进行了对比分析。实验结果表明,这两种方法相结合建立的非线性校正模型对预测集的预测精度有了很大的提高,这将对人体血糖浓度无创检测技术的研究具有实际应用价值。     

近红外光谱快速检测丙氨酸氨基转移酶

在探讨近红外光谱快速检测丙氨酸氨基转移酶的可行性过程中,首先对不同厚度(0.5,1,2和4mm)血液样品的近红外透射光谱进行了分析。发现全血样品0.5 mm厚时的近红外透射光谱更适合于进行光谱分析。进而采集了176个全血样品0.5 mm厚时的近红外光谱。对采集的光谱进行多元散射校正、二阶微分法光谱预处理后,采用逐步多元线性回归和偏最小二乘回归方法建立定量分析模型,预测了全血丙氨酸氨基转移酶的含量。结果表明:利用近红外光谱法测定丙氨酸氨基转移酶时,采用偏最小二乘回归方法建立的定标模型预测效果最好,定标相关系数、定标标准差和预测标准差的值分别为:0.98,2.42和7.22。     

Near-infrared spectroscopy for rapid evaluation of different processing products of Sophora japonica L.

To analyze the content of rutin in differently processed products of Sophora japonica L., a combination of near-infrared spectroscopy and chemometrics was used. Factors that affect near infrared spectroscopy modeling were investigated, and the optimal spectral conditions were determined. As a reference method, the rutin content was determined by high-performance liquid chromatography-diode array detection in 99 groups of products. Near-infrared spectra were collected by optimal near infrared acquisition conditions and the spectral features were enhanced by several preprocessing methods. Moreover, a near-infrared quantitative calibration model of rutin was established using partial least squares analysis. The root-mean-square standard error of cross validation and correlation coefficient (cross validation) were calculated as 0.800 and 0.9399, whereas the root-mean-square standard error of calibration and correlation coefficient (calibration) were 0.558 and 0.9709. This indicates that there is a good correlation between the predicted value and the actual measured value. Moreover, the root-mean-square standard error of prediction and correlation coefficient (validation) were calculated as 0.495 and 0.9785. And the ratio of the root-mean-square standard error of calibration and root-mean-square standard error of prediction was as small as 0.89 (≤1.2). The relative deviation between the predicted value and the reference value of the model was ?11.02 to 8.02%, and the ratio of performance to deviation was 4.70 (>3). Thus, these data indicate that the accuracy of the model used to predict the results was sufficient. In conclusion, the calibration model presented in this study provides satisfactory performance in a rapid, quantitative analysis of the rutin proportion by a simple, fast, and high accuracy of the prediction results.     

Selection of optimal combinations of inputs in a partial least squares model for prediction of soil organic matter

Abstract

Partial least squares model is widely used in estimation of soil physical and chemical parameters such as soil organic matter and moisture content, due to its advantages in dealing with collinearity of variables like hyperspectral reflectance. However, it is hard to determine optimal combination of partial least squares model input for soil organic matter prediction since there are lots of possibilities such as, different mathematical transformation of spectral reflectance, wavelength ranges, and spectral resolution. Laboratory hyperspectral reflectance of soils in Songnen plain were analyzed in this study, and the orthogonal experimental design method for deriving optimal combination of input variables for soil organic matter prediction models was introduced. For intercalating orthogonal experimental design table, five different levels which commonly used by researchers were assigned to factors. Results show that the optimal combination input for single black soil is using the derivative logarithmic reciprocal reflectance in the wavelength range selected by multiple stepwise regression at a spectral resolution of 5?nm (R²=?0.95, RMSE?=?0.21, and RPD?=?4.49), and different soils is using continuum removed in the wavelength range selected by MSR at a spectral resolution of 5?nm (R²?=?0.77, RMSE?=?0.74, and RPD?=?2.08). With optimal combination input, the partial least squares model prediction ability was evaluated as excellent for single black soil, possible for different soils. This study illustrates the orthogonal experimental design method can be an effective way to identify the optimal input variables of a partial least squares model for soil organic matter prediction, and multiple stepwise regression can be a preprocessing step to reduce hyperspectral data redundancy before using partial least squares to predict soil organic matter. Overall, this study provides a new approach for determining optimal input of partial least squares predicting model.     

Near-infrared spectroscopy to assess typhaneoside and isorhamnetin-3-O-glucoside in different processed products of pollen typhae

With the ever increasing importance of testing drug quality, rapid analytical methods are needed for supervision of Chinese herbal medicines. Near-infrared spectroscopy is one of the most powerful tools in quality assessment of Chinese herbal medicines. In this work, near-infrared spectroscopy was applied to develop a rapid method for quantitative determination of typhaneoside and isorhamnetin-3-O-glucoside in different processed products of Pollen Typhae. A total of 71 batches of samples were collected from different regions in China. After acquisition of near-infrared spectra, different pre-processing methods were compared, and a competitive adaptive reweighted sampling algorithm was used to perform the variable selection. Then a partial least squares regression algorithm was applied to build the quantitative models. The root mean square error of calibration, root mean square error of cross validation, and root mean square error of prediction were 0.0190, 0.0364, and 0.0158%, respectively, for a quantitative model of typhaneoside. The root mean square error of calibration, root mean square error of cross validation, and root mean square error of prediction were 0.0190, 0.0377, and 0.0170%, respectively, for a quantitative model of isorhamnetin-3-O-glucoside. Moreover, the relative prediction deviation values of both quantitative models were larger than 3, indicating good performance of the partial least squares (PLS) models. The results demonstrated that high accuracy prediction of typhaneoside and isorhamnetin-3-O-glucoside could be obtained by near-infrared spectroscopy, to allow an alternative method for quality assessment of different processed products of Pollen Typhae.     

堆叠监督自动编码器的近红外光谱建模

近红外光谱中包含了物质中有机分子含氢基团的特征信息,具有维度高、冗余大等特点。传统的基于浅层校正模型,比如主成分回归、偏最小二乘回归、人工神经网络、支持向量回归等,无法提取近红外光谱数据深层的信息。提出一种基于堆叠监督自动编码器的近红外光谱建模方法,不仅可以拟合光谱数据与理化值之间复杂的非线性关系,还可以提取数据深层的特征信息。首先通过对比不同的光谱预处理对模型预测结果的影响,选择最优的预处理方法,然后再使用相关系数法提取特征波段。将处理好的近红外光谱数据作为堆叠监督自动编码器的输入信号,利用理化值对多个监督自动编码器进行有监督的预训练;将多个经过预训练的监督自动编码器进行堆叠,得到堆叠监督自动编码器;将预训练的参数作为堆叠监督自动编码器的初始化参数,然后再利用理化值对堆叠监督自动编码器进行有监督的微调,最后得到模型的最优参数。分别利用玉米含水量和黄酒总酸含量等近红外数据集进行验证,建立了偏最小二乘回归预测模型、人工神经网络预测模型、堆叠自动编码器预测模型和堆叠监督自动编码器预测模型,验证了堆叠监督自动编码器建模的可行性;以预测均方根误差和预测相对分析误差两个指标对比分析了偏最小二乘回归、反向传播人工神经网络、堆叠自动编码器及堆叠监督自动编码器四种建模方法的评价指标。分析结果表明,采用该方法建立的模型,模型预测效果更好,玉米含水量数据集的两个评价指标达到了0.060 4和4.313;黄酒总酸含量数据的两个评价指标达到了0.120和4.227,均优于另外三种方法。     

奥林达夏橙叶片锌含量可见近红外光谱监测模型研究

以奥林达夏橙叶片粉末干样为对象,利用化学分析与可见近红外光谱技术相结合的方法,通过样品原始光谱的二阶微分及消噪(Noise)处理,运用偏最小二乘法(PLS)与交叉验证方法建立的Zn含量数学模型,其中使用Zn含量特征光谱400～500nm和1201～1300nm的组合波段建模,具有较好的预测能力,校正建模和预测模型的相关系数分别为0.9975和0.9920,交互验证预测均方根误差为0.5868。因此,利用可见近红外光谱技术,运用PLS及交叉验证方法,建立叶片Zn含量与特征波段的光谱校正模型,能快速定量检测柑桔叶片Zn含量。     

土壤碱解氮含量可见/近红外光谱预测模型优化

可见/近红外光谱技术是土壤成分检测的有效工具。波长筛选对可见/近红外模型土壤属性的预测精度有重要影响。以宁夏吴忠地区75个水稻土样为研究对象,利用可见/近红外光谱技术采集土壤样品光谱,采用SPXY （Sample set partitioning based on joint X-Y distance）方法选取了校正集和预测集样本,比较了分别采用Savitzky Golay平滑（SG smoothing）、多元散射校正（Multiple scatter correction,MSC）、标准正态变量变换（Standard normal variate,SNV）3种预处理方法对光谱数据处理后建立土壤碱解氮偏最小二乘法模型和原始光谱数据建模的效果。在此基础上,分别采用遗传算法（Genetic gorithms,GA）、连续投影算法（Successive projections algorithm,SPA）、竞争性自适应重加权算法（Competitive adaptive reweighted Sampling,CARS）、随机蛙跳（Random frog,RF）进行波长筛选,最后应用偏最小二乘法建立基于不同波长筛选方法的土壤碱解氮含量预测模型。研究表明,由于仪器性能稳定,样品的颗粒度比较小和均匀,本次实验原始光谱数据建模效果最好;各种波长筛选方法均可有效减少参与建模的波长数,且连续投影算法优于全谱建模,所选波长数仅为全谱波长数的1%,其预测决定系数（R²）、预测均方根误差和相对分析误差值分别为0.726,3.616,1.906。这表明连续投影算法可以有效筛选水稻土碱解氮敏感波段,为土壤碱解氮传感器开发提供技术支持。     

基于卡尔曼滤波的近红外光谱特征波长变量优选方法

将经典的卡尔曼滤波器与近红外光谱分析技术相结合,提出了一种新的特征波长变量选择方法——卡尔曼滤波法。分析了卡尔曼滤波器用于波长优选的原理,设计了波长选择算法并将其应用到大豆油脂酸价的近红外光谱检测中。首先利用偏最小二乘法(PLS)对油脂不同吸收波段建模,初步筛选出4 472～5 000 cm-1油脂酸价特征波段共132个波长点,然后进一步利用卡尔曼滤波器进行特征波长选择,从中优选出22个特征波长变量建立PLS校正模型,预测集决定系数R2、预测误差均方根RMSEP分别为0.970 8和0.125 4,与利用132个波长点建立的校正模型预测结果相当,而波长变量数减少到原来的16.67%。该波长变量选择算法是一种确定性的迭代过程,无复杂的参数设置和变量选择的随机性,物理意义明确。优选出少数对模型影响较大的特征波长变量以代替全谱建模,在简化模型的同时提高了模型的稳健性,为开发专用油脂近红外光谱分析仪器提供了重要参考依据。     

Selection of near infrared spectral information based on section combination moving window partial least squares北大核心CSCD

Selection of near infrared spectral information is research focus on the application of NIR, which enable to simplify the model and improve accuracy of prediction. Aiming at selecting optimal modeling spectral width of near infrared spectroscopy, section combination moving window partial least squares (SCMWPLS) is proposed in this paper. This method selects continuous modeling screened interval.by continuously varying size of moving window and cross validation. Taking Glucose solution near-infrared spectroscopy as test specimen, near infrared prediction models are established respectively by proposed method, and traditional interval partial least squares (IPLS) and moving window partial least squares (MWPLS). Comparing proposed method with two traditional methods, squares prediction RMSE is decreased by 44% and 25% respectively. © 2017, Editorial Board, Journal of Applied Optics. All right reserved.     

基于小波压缩的马铃薯全粉还原糖近红外光谱检测研究

为探讨小波压缩算法结合近红外光谱技术在马铃薯全粉还原糖含量检测中的可行性,采用傅里叶变换近红外光谱仪采集了250份马铃薯全粉样品的近红外光谱。分别优化了消失矩、小波系数和主成分因子数,优化结果为10,100和20。基于db小波函数将1 501个马铃薯全粉的近红外光谱变量压缩成100个小波系数。分别以1 501个光谱变量和100个小波系数为变量分别建立了偏最小二乘(PLS)校正模型。以62个未参与建模的样品作为预测集,考察模型的预测能力。经比较,小波压缩结合PLS的校正模型预测结果最优,模型预测相关系数为0.98,预测均方根误差为0.181%。实验结果表明小波压缩算法结合近红外光谱技术有效地保留了有效光谱信息,实现了光谱数据降维,简化了马铃薯全粉还原糖PLS校正模型,提高了模型的预测能力。     

波长漂移对近红外光谱PLSR分析模型的影响

在近红外光谱分析过程中,单台仪器在不同时间的波长变化及多台仪器间的波长一致与否会对化学计量学定标模型的校正及传递效果产生影响,上述问题可以统一为波长漂移对定标模型的影响。以分析小麦粉中粗蛋白含量为例,首先结合不同谱区光谱数据,利用偏最小二乘回归（PLSR）方法建立了两个定标模型。再由计算机生成不同类型、不同幅度的波长漂移信息,并叠加至验证集样品光谱中,使新光谱相对于定标集光谱产生波长漂移信息。通过考察原定标模型对新光谱的预测与校正情况,研究了波长漂移对PLSR定标模型的影响。结果表明：相对于定标集样品光谱,验证集光谱中无波长漂移信息时,模型的预测标准差（RMSEP）不超过0.3%,预测相关系数不小于0.98;验证集样品光谱在不同波长处的波长漂移信息为一恒定值时,模型的RMSEP会随波长漂移幅度的增大而增大,波长漂移量为-32 cm-1时对应RMSEP为3.69%,预测相关系数变化不大;当验证集样品光谱在不同波长处的波长漂移信息随机变化时,基于长波区光谱所得原始模型的预测结果几乎不受影响;当含有不同波长漂移信息的一系列样品光谱加入到定标集对长波区PLSR分析模型进行校正时,校正后模型的RMSEP为0.3%,几乎不受波长漂移信息的影响,但模型的回归因子数从3显著增大到8,其稳健性变弱;总的来说,当仪器存在波长漂移且幅度不大时,模型预测相关系数几乎不受影响,可通过对预测结果的校正来改善RMSEP,以保证分析结果的准确性。该研究为确定仪器设计参数及分析方法的操作规程,提高近红外光谱分析结果的可靠性提供了实验依据。     

基于CART回归树的LIBS特征变量选择方法研究

激光诱导击穿光谱技术（LIBS）用于检测时,由于谱线多且复杂,存在许多冗余的信息,这些都会对定量分析造成影响。因此,提取有效的特征变量在LIBS的定量分析中具有非常重要的意义。对CaCl₂溶液中的Ca元素进行光谱特征选择方法分析,对比单变量模型、偏最小二乘回归和CART回归树定标模型的准确度和稳定性。针对水体表面的波动性较大,光谱稳定性差,同时光谱受基体效应和自吸收效应影响等问题,首先采用单变量模型得到的拟合系数（R2）仅有0.933 2,训练均方根误差（RMSEC）、预测均方根误差(RMSEP)和平均相对误差(ARE)分别为0.019 2 Wt%,0.017 7 Wt%和11.604%。经偏最小二乘回归优化后,模型R2提高到0.975 3,RMSEC,RMSEP和ARE分别降低到0.010 8 Wt%,0.013 Wt%和7.49%。为了进一步提高定量分析的准确度,建立CART回归树定标模型。该方法在构建树模型时,通过平方误差最小化准则,从复杂的光谱信息中选取最优的特征变量组合做分类决策,从而建立Ca元素的定标曲线。通过CART回归树的变量选择,特征变量个数从100个减少到6个,变量的压缩率达到了94%,显著降低了无关谱线的干扰,回归树模型的相关系数R2,RMSEC,RMSEP和ARE分别为0.997 5,0.003 5 Wt%,0.006 1 Wt%和2.500%。相较于传统的单变量模型与偏最小二乘回归,CART回归树模型具有更高的精度、更小的误差。通过对特征变量的有效筛选,剔除无关信号的干扰,显著降低了基体效应和自吸收效应对LIBS定量分析的影响,提高了定量分析的准确度和稳定性。     

激光诱导击穿光谱联合UVE变量优选检测大豆油中的铬含量

利用激光诱导击穿光谱(LIBS)技术对大豆油中的重金属Cr进行检测研究。以松木木片对重金属Cr进行富集,采用AvaSpec双通道高精度光谱仪在206.28~481.77 nm波段范围内采集松木木片样本的LIBS光谱,利用无信息变量消除(UVE)方法筛选与重金属Cr相关的波长变量,应用偏最小二乘(PLS)回归建立大豆油中重金属Cr的定标模型,并与单变量及全波段PLS定标模型进行比较。结果表明,相比单变量及全波段PLS定标模型,UVE-PLS定标模型的性能更优,其相关系数、校正均方根误差、交互验证均方根误差及预测均方根误差分别为0.990,0.045,0.050及0.054 mg·g-1。经UVE变量筛选后,UVE-PLS定标模型所用的波长变量数仅为全波段PLS的2%。由此可见,UVE是一种有效的波长变量筛选方法,能有效筛选出与重金属Cr相关的波长变量。