Hyperspectral estimation of soil organic matter based on different spectral preprocessing techniques

In this study, the potentiality of visible and near-infrared reflectance spectroscopy to estimate soil organic matter was assessed. Six preprocessing methods were implemented to process the original spectra. The partial least-squares regression approach was also applied to construct predictive models and evaluate the optimal spectral preprocessing method. The significant wavelengths of soil organic matter were determined by using the correlation analysis and the partial least-squares regression analysis. The results were: (i) visible and near-infrared reflectance spectroscopy was proved to be an ideal approach in the soil organic matter estimation; (ii) different preprocessed spectra could improve their correlation with soil organic matter; the combination of first-order derivative and Savitzky–Golay smoothing method outperformed other preprocessing methods; (iii) the soil organic matter predictive models based on spectra processed by derivatives and Savitzky–Golay smoothing together presented a satisfactory accuracy, yielding the determination coefficient and root mean square error values of 0.986 and 0.077, respectively, for first-order derivative; and 0.973 and 0.105, respectively, for second-order derivative. The combination of first-order derivative and Savitzky–Golay smoothing was ultimately recommended the preferable preprocessing method; and (iv) the wavelengths of 417, 1853, 1000, and 2412?nm were determined as the significant wavelengths associated with soil organic matter. The study will provide a reference for the site specific management of agricultural inputs by using the visible and near-infrared reflectance spectroscopy technology.     

一种基于无监督主动学习的苹果品质光谱无损检测模型构建方法

利用光谱技术实现农产品、食品品质无损检测的实质是建立样本光谱信息与样本品质参数之间的机器学习模型。为了获得具有良好泛化性能的机器学习模型,通常需要大量的标记样本,然而,获取样本的光谱信息相对容易,但标注样本品质参数的过程往往涉及到大量的时间和经济成本,并且具有破坏性。主动学习是一种减少训练集有标记样本数量的方法,通过选择最有价值的样本进行标记,而不是随机选择。因此,主动学习能够控制向训练集添加哪些样本,模型不再是被动地接受用于建模的样本。在分类任务中已经提出较多关于主动学习的算法,但回归任务中的研究却相对较少,且现有的用于回归任务的主动学习算法大多是有监督的,即需要少量有标记样本训练初始模型。本文提出了一种基于无监督主动学习方法的训练样本选择策略。该方法首先通过层次凝聚聚类对无标记（标准值）光谱数据集进行多样性划分,获得不同的聚类簇;然后通过局部线性重建算法在每个聚类簇中选择最具代表性的样本构成训练样本集,最后基于训练集构建模型。利用两个年份三个品种苹果的近红外光谱数据,构建了其可溶性固形物含量和硬度的偏最小二乘预测模型,用于验证所提出方法的有效性。实验结果表明：所提出的方法要优于已有的样本选择策略,可以有效地提高模型精度,减少在模型训练中的破坏性理化实验。同时,与随机采样（RS）、Kennard-Stone算法（KS）、光谱-理化值共生距离算法（SPXY）这三种光谱领域常用的样本选择算法相比,该研究所提出的方法表现出了最佳的性能, 基于所提出的无监督主动学习算法选取200个样本作为训练集所建立的可溶性固形物含量预测模型的预测均方根误差相对于其他三种算法降低了2.0%~13.2%,硬度预测模型的预测均方根误差相对降低了1.2%~15.7%。     

可见光-近红外、中红外光谱的土壤有机质组分反演

土壤有机质是土壤肥力的物质基础,其含量的高低是评价土壤肥力的重要标志。土壤有机质组分根据其溶解性可分为胡敏素(HM)、胡敏酸(HA)、富里酸(FA),不同组分的肥力特性差异显著,因此,土壤有机质组分数据可更加全面、客观的反映土壤肥力状况。传统土壤土壤有机质及组分的测定工序繁杂,效率低下且时效性差,大量研究表明高光谱技术能有效提高土壤属性的检测效率并降低测试成本,但关于可见光-近红外、中红外光谱检测土壤有机质组分的报道鲜见。为了探索中红外光谱及可见光-近红外-中红外组合光谱对土壤有机质组分检测的可行性,并对比有机质单一光谱模型与有机质不同组分的组合光谱模型的预测精度,以南疆地区农田土壤为例,在阿克苏及和田地区共采集93个土样,进行有机质、胡敏素、胡敏酸、富里酸含量及光谱数据的测定。其次,利用可见-近红外(VNIR)、中红外(MIR)及其组合光谱(VNIR-MIR)三种光谱数据集,采用偏最小二乘(PLSR)、支持向量机(SVM)、随机森林(RF)三种建模方式对土壤有机质、胡敏素、胡敏酸、富里酸含量进行组合模型分析预测。结果表明：(1)土壤有机质及各组分均与光谱反射率有较好的相关性,土壤有机质及组分在MIR谱段的特征波段数量明显多于VNIR谱段。(2)有机质最优预测模型的模式为VNIR-MIR-RF,该模型的决定系数R2为0.90;胡敏素与胡敏酸最优预测模型的模式均为VNIR-RF模型,R2均为0.92;富里酸最优预测模型的模式为MIR-RF模型,R2为0.94。(3) 基于胡敏素、胡敏酸和富里酸的有机质组合光谱模型的预测精度明显高于有机质单一光谱模型,两种模型的R2分别为0.93和0.90。实现了土壤有机质组分的高效快速反演,且基于有机质组分的组合模型提高了土壤有机质预测精度,为南疆地区大尺度土壤肥力的鉴定与精准施肥提供重要的参考价值。     

水稻土可见-近红外-中红外光谱特性与有机质预测研究

土壤有机质是农田肥力评估的重要指标,要实现快速获取大面积土壤有机质的含量需要建立高效、稳健的预测模型。光谱技术能够快速诊断土壤有机质,以水稻土为例,从校正样本选择方法的对比,研究了可见-近红外、中红外和可见-近红外-中红外三种不同波段光谱对土壤有机质的预测能力。可见-近红外和中红外区域的光谱反射率转换成吸收率后通过Savitzky-Golay平滑法去噪,通过三种校正样本选择方法建立相应的偏最小二乘回归预测模型。通过Rank-KS法建立的三种波段的有机质预测模型均优于Rank法和KS法,中红外波段光谱的模型预测能力强于可见-近红外和可见-近红外-中红外波段的预测模型,基于Rank-KS法建立的中红外波段有机质预测模型取得了最好的预测效果,RMSE_P仅为3.25 g·kg-1,RPD达到4.24,依据VIP得分筛选出可见-近红外和中红外波段的水稻土有机质重要建模波段。因此,中红外光谱建模技术能够对水稻土有机质进行快速定量分析,Rank-KS法可提高模型的预测能力,为今后农田肥力评价和科学施肥提供技术支持。     

西范坪矿区土壤铜元素的高光谱响应与反演模型研究

为解决传统的土壤地球化学测量方法成本高、效率低等问题,研究了利用可见-近红外光谱技术检测土壤重金属含量的简易方法。研究对西范坪矿区土壤反射光谱进行微分、连续统去除等六种变换,利用逐步回归法和皮尔逊相关系数选出对土壤铜含量敏感的特征波段,组成综合特征变量集,再应用不同的特征变量选取方法和参数建立估算模型并检验。结果表明：不同的光谱变换方法对土壤铜含量信息提取能力不同,每种光谱变换都对应特定的敏感波谱区间;基于综合光谱变换信息建立的土壤铜含量反演模型精度优于基于单种光谱变换信息建立的模型;利用综合光谱变换信息建立土壤铜含量反演模型,后向剔除法优于前向引入法和逐步回归法,当Removal取0.20时得到最优回归模型,其模型决定系数R2和预测模型决定系数R2_pre分别达到了0.851和0.830,建模均方根误差RMSEC和预测均方根误差RMSEP分别为0.349和0.468 mg·kg-1,能较好地检测土壤铜含量,同时为其他土壤重金属元素的光谱检测提供了思路。     

Optimized prediction of sugar content in ‘Snow’ pear using near-infrared diffuse reflectance spectroscopy combined with chemometrics

In near-infrared spectroscopy applications, the original spectra often contain redundant information, which will seriously affect the performance of chemometric models. Therefore, preprocessing, effective wavelengths selection, and appropriate regression models are essential. The objective of this study was to optimize the nondestructive determination multivariate calibration model of sugar content in ‘Snow’ pears, using near-infrared diffuse reflectance spectroscopy combined with chemometrics. All data (sugar content reference values and spectra data) from three measuring positions (P1, P2, and P3, marked around the pear’s equator at angular distances of approximately 120°) were divided into four grouped datasets, namely Set-1 (P1), Set-2 (P2), Set-3 (P3), and Set-4 (average of the three measuring positions). All subsequent optimized processes were performed based on each grouped dataset. First, different preprocessing methods were tested and an optimal method was determined. Then, synergy interval partial least squares and synergy interval partial least squares-competitive adaptive reweighted sampling were applied to select effective regions and effective wavelengths from all wavelengths, respectively, and partial least squares regression models were established and analyzed. In addition, support vector regression models were also established for comparative study. After comprehensive analysis of prediction accuracy and model complexity, the partial least squares regression model based on the 16 selected effective wavelengths for Set-4 was optimal, with the correlation coefficient for prediction, root-mean-square error of prediction, and residual predictive deviation of 0.9701, 0.2311, and 4.12, respectively. The results indicated that with these optimized processes, the multivariate calibration model of sugar content in ‘Snow’ pears was effectively optimized for each dataset. In addition, it is concluded that partial least squares regression was superior to support vector regression in this study, although some other researches had found different results in related fields.     

Rapid evaluation of soil fertility in tea plantation based on near-infrared spectroscopy

Abstract

An efficient soil analysis technique was developed to monitor soil fertility and perform precise soil management in tea plantations. In this study, near-infrared spectroscopy combined with chemometric methods was utilized to determine the organic matter and total nitrogen content and evaluate fertility of tea plantation soils. First, photometric precision and subtractive spectroscopy were used as indicators in identifying optimal sample preparation condition. Spectral reproducibility reached an optimum with powder particle size of 100 mesh (0.149?mm). Second, after comparing the combinations of the partial least squares method with three different characteristic wavelength extraction methods, the genetic algorithm and competitive adaptive reweighted sampling quantitative discrimination models were determined to be optimal for organic matter and total nitrogen contents, with prediction correlation coefficients of 0.9102 and 0.8763, respectively. Third, classification models for soil fertility level using linear discriminant analysis, support vector machine, and extreme learning machine were established based on a full spectrum and successive projections algorithm separately. The successive projections algorithm-extreme learning machine model was deemed superior with a correct classification rate of 84.38%. Our findings demonstrate that the proposed near-infrared spectroscopy calibration models successfully achieve the nondestructive and rapid evaluation of organic matter and total nitrogen contents, as well as the classification of soil fertility levels, in tea plantation soils. The results provide a basis for the development of internet-of-things sensors in the construction of a high-yield and high-quality tea plantation.     

不同土壤类型的有机质含量的可见-近红外光谱检测模型传递方法研究

利用可见-近红外光谱分析技术可以准确快速的获取土壤养分含量,但不同类型土壤间养分含量校正模型的普适性是亟待解决的关键问题。为提高有机质含量光谱校正模型在多类型土壤之间的普适性和农田在线检测有机质含量速度,利用美国M107B区66个样品建立基于可见-近红外光谱的土壤有机质含量的粒子群-最小二乘支持向量机（PSO-LSSVM）校正模型,预测M107B区的23个验证集样品的决定系数R2=0.859,相对分析误差RPD=2.660;将M107B区89个土壤样品作为校正集建模后对N116B区20个验证集样品的有机质含量预测,预测R2=0.562,预测RPD=0.952,模型的预测R2和预测RPD分别降低34.6%和64.2%,表明M107B区土壤有机质含量的可见-近红外光谱校正模型直接用于N116B区时,预测精度显著降低;将N116B区部分土壤样品加入到M107B区样品集后重新建模,并预测N116B区20个验证集样品的有机质含量,当加入的N116B区土壤样品数量达到35以上,预测R2>0.80,预测RPD>2.0;加入到校正集的N116B区土壤样品数量从0增加到50,模型预测R2从0.562增加到0.811,预测RPD从0.952增加到2.274,精度逐渐提高。结果表明,在M107B区校正模型中加入N116B区部分土壤样品建模,能够有效提高M107B区土壤校正模型对N116B区土壤有机质含量的预测精度;加入的N116B区土壤样品数量达到50以上,模型预测性能趋于稳定,预测精度达到实用要求,成功将M107B区土壤有机质含量校正模型传递给N116B区土壤;优先选择与M107B区土壤样品的有机质含量或光谱曲线差异较大的N116B区土壤样品参与建模,可有效避免模型传递时模型性能出现突变。提出的方法能够有效提高M107B区土壤的有机质校正模型对N116B区土壤的预测精度,为基于可见-近红外光谱的农田土壤有机质含量实时检测提供一种新的经济可行的模型传递方法,为提高多类型土壤的有机质含量检测模型的普适性提供一种有效的解决方案。     

南疆农田土壤全氮含量的中红外光谱反演模型

快速准确监测农田土壤全氮含量,可显著提高土壤肥力诊断与评价工作的效率。传统测定土壤全氮的方法存在耗时费力、成本高、环境污染等缺点,而基于光谱学原理的土壤全氮定量方法克服了传统测量的劣势。中红外(MIR)光谱相较于可见光-近红外(VNIR)光谱而言,具有更多的波段数和信息量,如何利用中红外光谱监测土壤全氮含量是具有重要应用前景的研究课题。为了探索中红外光谱对土壤全氮监测的可行性,以新疆南疆地区采集的246个农田土样为研究对象,以室内测定的全氮含量和中红外光谱反射率数据为数据源,分析了不同全氮含量土样的中红外光谱特征差异,以主成分分析法(PCA)和连续投影算法(SPA)对光谱数据进行降维,然后采用偏最小二乘回归(PLSR)、支持向量机(SVM)、随机森林(RF)和反向传播神经网络(BPNN)四种建模方法分别构建基于全波段和降维数据的土壤全氮含量定量反演模型。研究结果表明：(1)土壤在中红外波段光谱反射率随全氮含量的增加而增加,在3 620, 2 520, 1 620和1 420 cm^-1附近存在明显的吸收谷;将中红外光谱数据进行最大值归一化处理后,可明显提高土壤光谱反...     

土壤主要养分近红外光谱分析及其测量系统

土壤是农业生产的基础,采用近红外光谱技术实现对土壤养分的快速分析,研制分立波长型近红外土壤养分测量系统,指导农业生产过程,有助于改变现有农业生产的粗放经营状态.首先,使用FOSSXDS近红外光谱分析仪对85份东北土壤样品采集光谱,采用相关系数谱及连续投影法等化学计量学算法分析土壤的近红外光谱,并优选出总氮和有机质的特征...     

利用近红外及中红外融合技术对小麦产地和烘干程度的同时鉴别

小麦是制作馒头的主要原料之一,小麦中水、蛋白质、淀粉会因产地以及烘干程度的差异而不同,进而影响到加工成馒头的品质。所以实现对小麦产地和烘干程度的快速鉴别就显得尤为重要。感官评定是鉴别小麦产地和烘干程度常用的方法,对比感官评定,光谱分析可以识别样品中的分子结构等信息。基于此,尝试利用近红外和中红外光谱融合技术实现对不同产地和不同烘干程度的小麦同时鉴别。首先选取了两个不同产地的小麦,再利用微波干燥法对两个不同产地的小麦做烘干预处理,使烘干的小麦水含量为12%±0.5%,原麦水含量为18%±0.5%。分别标记为原麦A,烘干A,原麦B,烘干B,再将小麦研磨成粉末,过100目筛网筛选后,置于自封袋中备用。随后分别采集四种小麦样品的近红外和中红外光谱信息,在Matlab 7.10的环境下使用标准正态变量变换(standard normal variable transformation, SNVT)对采集到的原始光谱数据进行预处理,利用主成分分析对预处理后的数据进行降维处理,再结合线性判别分析（linear discriminant analysis,LDA）和支持向量机（support vector machine, SVM）分别建立小麦近红外、中红外光谱数据识别模型。另外利用联合区间偏最小二乘法（synergy interval partial least square, SiPLS）筛选出利用标准正态变量变换(SNVT)预处理后的小麦近红外和中红外光谱数据特征光谱区间,将筛选出的近红外和中红外光谱数据特征光谱区间融合后再结合线性判别分析（LDA）和支持向量机（SVM）建立小麦融合光谱信息的识别模型。然后比较同种光谱数据下利用线性判别分析（LDA）和支持向量机（SVM）建立的小麦识别模型识别率、比较同种建模方法下近红外和中红外光谱数据建立小麦识别模型识别率、比较同种建模方法下光谱数据融合和单一光谱数据建立小麦识别模型识别率。结果表明,同种光谱分析方法,利用SVM建立的四种小麦识别模型识别率高于利用LDA建立的小麦识别模型识别率。同种建模方法,近红外光谱数据建立的小麦识别模型识别率优于中红外光谱数据建立的小麦识别模型识别率。而在同种建模方法下,利用SiPLS筛选出近红外和中红外光谱数据的特征光谱区间数据融合后建立小麦识别模型识别率最高,光谱数据融合后结合LDA建立的小麦识别模型校正集识别率为98.75%,预测集识别率为97.50%;而将此选择的变量结合SVM建立的小麦识别模型的校正集和预测集识别率都达到100.0%。对比利用单一光谱数据建立的小麦识别模型识别率,光谱数据融合之后建立的小麦识别模型识别率得到显著提高,该研究从纵向和横向上全面地比较了光谱数据建立的小麦模型识别率,结果可为更准确地运用光谱融合技术建立小麦产地以及烘干程度识别模型提供参考。     

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。     

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...     

Rapid nondestructive evaluation of duck meat pH and color using visible and near-infrared reflectance spectroscopy

In the present study, visible and near-infrared reflectance spectroscopy were applied to predict quality attributes of duck breast meat. The real color (expressed as lightness, redness, and yellowness) and pH of duck samples were recorded using traditional contact methods and then modeled with their corresponding spectral data by partial least square regression. Before the establishment of prediction models, three different spectral preprocessing methods including first derivative, standard normalized variate, and Savitzky–Golay smoothing were used. Compared to the models obtained from original spectral data set, the predictive capabilities of models based on the spectra after preprocessing were improved effectively. As a result, the determination coefficient of calibration set and prediction set of the best models for lightness, redness, yellowness, and pH parameters were 0.96 and 0.85, 0.94 and 0.94, 0.96 and 0.94, 0.81 and 0.76, respectively. Results demonstrated that visible and near-infrared spectroscopy can become a useful tool for rapid prediction of duck color and pH quality attributes.     

Adulteration of sesame oil with corn oil detected by use of two-dimensional infrared correlation spectroscopy and multivariate calibration

An innovative methodology was developed to detect adulteration of sesame oil with corn oil based on two-dimensional mid-infrared correlation spectroscopy with multivariate calibration. Forty pure sesame oils and 40 adulterated sesame oils with corn oil were prepared and the infrared absorption spectra were measured at room temperature, respectively. The synchronous two-dimensional mid-infrared correlation spectra were calculated to develop multivariate calibration models for adulteration of sesame oil with corn oil. The results showed the higher classification accuracy of 96.3% for the prediction set using two-dimensional mid-infrared correlation spectra and N-way partial least square discriminant analysis, versus 88.9% using traditional one-dimensional mid-infrared spectra and partial least squares discriminant analysis. Also, the multivariate calibration models were developed for quantitative analysis of sesame oil adulteration with corn oil. The root mean square error of prediction was 0.98% v/v using two-dimensional mid-infrared correlation spectra and N-PLS, and 1.15% v/v using traditional one-dimensional mid-infrared spectra and PLS. The results of our analyses indicated that the proposed method could provide better predictive results than traditional one-dimensional mid-infrared spectra and multivariate calibration.     

应用近红外高光谱成像预测三文鱼肉的水分含量

应用近红外高光谱成像技术实现三文鱼肉水分含量的快速无损检测。采集来自不同部位的三文鱼肉共90个样本的高光谱图像,提取样本感兴趣区域(ROI)的平均光谱。随机取60个样本作为建模集,其余30个样本作为预测集。分别采用偏最小二乘回归(PLSR)和最小二乘支持向量机(LS-SVM)对全波段和水分含量建立相关性模型,并对预测集样本的水分含量进行预测。再用一种新的变量提取方法random frog选择特征波长,并基于特征波长分别建立水分检测的PLSR和LS-SVM模型。特征波长模型的预测精度虽然稍逊于全波段模型,但是仅用12个变量代替了全波段的151个变量,大大简化了模型,更便于实际应用。PLSR和LS-SVM特征波长模型的预测相关系数(R_p)分别为0.92和0.93,预测均方根误差(RMSEP)分别为1.31%和1.18%,取得了满意的结果。研究表明,近红外高光谱成像与化学计量学方法结合可以准确预测三文鱼肉的水分含量,为鱼肉品质的快速监测提供重要的参考。     

Feasibility of estimating copper content using visible and near-infrared spectra of bedrock,weathered rock and soil - a case study of the Baogutu stock II,Xinjiang, China

Geochemical anomalies are important symbols for prospecting. To determine a new method for identifying geochemical anomalies rapidly, this study examines the possibility of using visible and near-infrared spectra of bedrock, weathered rock, and soil to assess copper content in the Baogutu stock II, Xinjiang, China. Partial least squares regression was employed to study the relationship between visible and near-infrared spectra and the ore-forming element contents of collected bedrock samples, weathered rock samples, and soil samples. More specifically, our study aims to achieve three objectives. First, to explore the correlations between the contents of different elements and the correlations between the visible and near-infrared spectra and the ore-forming element contents for the three sample types. Second, to evaluate the performance of the reflectance-based partial least squares regression model. The models for iron in bedrock samples and copper in soil samples were chosen for estimating copper content. Third, to assess the effectiveness of data transformation methods in the partial least squares regression model. In our case study, the square root-based models for iron in bedrock samples and copper in soil samples with coefficients of determination for prediction are 0.614 and 0.409 respectively, maximized the model performance. Therefore, it is evident that visible and near-infrared spectra can be an alternative method for estimating ore-forming element contents based on bedrock, weathered rock, and soil samples.     

基于全谱数据挖掘技术的土壤有机质高光谱预测建模研究

可见/近红外高光谱技术与建模方法是当前土壤近地传感器研究领域的重要方向,可应用于土壤养分信息的快速获取和农田作物的精确施肥管理。以浙江省水稻土为研究对象,利用以非线性模型为核心的数据挖掘技术,包括随机森林、支持向量机、人工神经网络等方法分别建立了不同建模集和验证集的原始光谱与有机质含量的估测模型。结果表明：研究比较的1∶1,3∶1和全部样本建模并全部验证的三种样本模式划分对建模的结果有一定的影响。相较于目前常用的偏最小二乘回归(PLSR)建模方法而言,非线性模型RF和SVM也取得了较好的建模精度,三种模式下其RDP值均大于1.4。特别是采用SVM建模方法所得模型具有很好的预测能力,模式二下其RDP值达到2.16。同时引入ANN方法改进建立的PLSR-ANN方法显著提高了PLSR的模型预测能力。     

应用近红外光谱对低碳数脂肪酸含量预测

应用近红外光谱技术结合支持向量机回归(support vector machine regression, SVR)方法测量食用植物油脂低碳数脂肪酸(C≤14)含量。使用SupNIR-5700近红外光谱仪采集58个样品的近红外光谱图,通过偏最小二乘(partial least square, PLS)算法剔除奇异样品。选择其中具有代表性的52个样品进行主成分分析(principal component analysis, PCA),选取径向基(radial basis function, RBF)核函数建立支持向量机回归模型,并对光谱预处理方法和参数寻优方法进行了详细的分析和讨论。实验表明,经过粒子群算法(particle swarm optimization, PSO)优化后模型的性能都有所提高,泛化能力更强,预测的准确度和稳健性更好;其中预处理方法2经过PSO优化寻优后的参数C=2.085, γ=22.20时,预测集和校正集相关系数(correlation coefficient, r)分别达到了0.998 0和0.925 8,均方根误差(root mean square error, MSE)分别为0.000 4和0.014 3。研究结果表明,应用近红外光谱结合PSO-SVR方法进行食用植物油脂低碳数脂肪酸含量快速、准确的预测是可行的。     

近红外光谱烟叶质量等级快速检测与等级特征分析

烤烟等级质量对配方设计和卷烟工业生产的稳定起着重要的作用.采用传统外观分级法对2018年全国40个地级市产地的768份烤烟烟叶样品进行分类定级,包括7个不同烟叶质量等级.应用近红外光谱建立烟叶质量等级预测模型,分析不同等级烟叶化学基团及相关成分的近红外吸收光谱特征.结果表明:不分产区建立全国烟叶等级预测模型,建模集与预...