Author Summary: Transfer of calibrations between instruments is a key issue to use the value of a calibration over multiple units. Transfer is relatively easy between spectrometers of the same type but can be problematic between different instrument models. Two hundred and seventy samples of wheat from Northern Italy were scanned using a Corona Extreme and an Aurora handheld NIR. Samples (n = 46) from three different locations were removed from the original dataset and used for external validation. The PLS calibrations performances were satisfactory, with SECV for Moisture of 0.09 % and 0.13 % and for Protein of 0.28 % and 0.45 %, respectively for Aurora handheld NIR and Corona Extreme. Performance of validation (SEP) within instrument was of 0.07 % and 0.11 % for Moisture and of 0.27 % and 0.37 % for Protein, for the handheld and the process instrument, respectively. When the same calibrations were used to predict samples across instruments, the SEP was of 0.08 % and 0.19 % for Moisture and of 0.34 % and 0.47 % for Protein, for Corona Extreme predicting Aurora handheld NIR and vice versa, respectively. Both instruments can accurately predict the parameters of interest on wheat and could use the same calibration avoiding time-consuming standardization procedures.
{"title":"Transfer of grain calibrations between a handheld and a process instrument","authors":"F. Benozzo, P. Berzaghi","doi":"10.1255/NIR2017.011","DOIUrl":"https://doi.org/10.1255/NIR2017.011","url":null,"abstract":"Author Summary: Transfer of calibrations between instruments is a key issue to use the value of a calibration over multiple units. Transfer is relatively easy between spectrometers of the same type but can be problematic between different instrument models. Two hundred and seventy samples of wheat from Northern Italy were scanned using a Corona Extreme and an Aurora handheld NIR. Samples (n = 46) from three different locations were removed from the original dataset and used for external validation. The PLS calibrations performances were satisfactory, with SECV for Moisture of 0.09 % and 0.13 % and for Protein of 0.28 % and 0.45 %, respectively for Aurora handheld NIR and Corona Extreme. Performance of validation (SEP) within instrument was of 0.07 % and 0.11 % for Moisture and of 0.27 % and 0.37 % for Protein, for the handheld and the process instrument, respectively. When the same calibrations were used to predict samples across instruments, the SEP was of 0.08 % and 0.19 % for Moisture and of 0.34 % and 0.47 % for Protein, for Corona Extreme predicting Aurora handheld NIR and vice versa, respectively. Both instruments can accurately predict the parameters of interest on wheat and could use the same calibration avoiding time-consuming standardization procedures.","PeriodicalId":20429,"journal":{"name":"Proceedings of the 18th International Conference on Near Infrared Spectroscopy","volume":"277 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"83066695","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Author Summary: Transfer methods were compared for the porting of partial least squares models for intact mango dry matter content between short wave near infrared silicon photodiode array instruments. Methods included bias adjustment using average difference spectrum, new pixel-to-wavelength assignments, piecewise direct standardisation (PDS), global models, model updating (MU) and combinations of these. Best results (R2 > 0.84 and bias < 0.2) were obtained by PDS using the same variety of fruit in calibration and transfer sets. The use of an apple spectra transfer set was also successful, if the wavelength accuracy of the slave unit(s) is satisfactory. Alternatively, a field practical solution that gave acceptable prediction results involved development of a global model across units or model updating by inclusion of spectra of the new population, using reference values estimated using the master unit.
{"title":"Calibration transfer between short wave near infrared photodiode array instruments","authors":"C. Hayes, K. Walsh, R. Lerud","doi":"10.1255/NIR2017.071","DOIUrl":"https://doi.org/10.1255/NIR2017.071","url":null,"abstract":"Author Summary: Transfer methods were compared for the porting of partial least squares models for intact mango dry matter content between short wave near infrared silicon photodiode array instruments. Methods included bias adjustment using average difference spectrum, new pixel-to-wavelength assignments, piecewise direct standardisation (PDS), global models, model updating (MU) and combinations of these. Best results (R2 > 0.84 and bias < 0.2) were obtained by PDS using the same variety of fruit in calibration and transfer sets. The use of an apple spectra transfer set was also successful, if the wavelength accuracy of the slave unit(s) is satisfactory. Alternatively, a field practical solution that gave acceptable prediction results involved development of a global model across units or model updating by inclusion of spectra of the new population, using reference values estimated using the master unit.","PeriodicalId":20429,"journal":{"name":"Proceedings of the 18th International Conference on Near Infrared Spectroscopy","volume":"1993 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82401120","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
O. Minet, V. Baeten, B. Lecler, P. Dardenne, J. Pierna
Author Summary: The purpose of this study was to evaluate two different locally based regression methods (LOCAL and Local Calibration by Customized Radii Selection) and compare their performance to the classical global PLS for large NIR data. The data used in this study came from two inter-laboratory studies for wheat grain analysis organized in 2016 in the framework of the REQUASUD network. The results showed that improved predictions in terms of prediction errors can be obtained using local approaches compared to the classical global PLS. Moreover, the study highlighted clear differences between inter-laboratory studies and participating laboratories, which were even more evident when working with local procedures.
作者摘要:本研究的目的是评估两种不同的基于局部的回归方法(LOCAL和LOCAL Calibration by Customized Radii Selection)，并将它们与经典的全局PLS在大近红外数据中的表现进行比较。本研究中使用的数据来自2016年在REQUASUD网络框架下组织的两项小麦籽粒分析实验室间研究。结果表明，与经典的全局PLS相比，使用局部方法可以获得预测误差方面的改进预测。此外，该研究强调了实验室间研究和参与实验室之间的明显差异，这在使用局部程序时更为明显。
{"title":"Local vs global methods applied to large near infrared databases covering high variability","authors":"O. Minet, V. Baeten, B. Lecler, P. Dardenne, J. Pierna","doi":"10.1255/NIR2017.045","DOIUrl":"https://doi.org/10.1255/NIR2017.045","url":null,"abstract":"Author Summary: The purpose of this study was to evaluate two different locally based regression methods (LOCAL and Local Calibration by Customized Radii Selection) and compare their performance to the classical global PLS for large NIR data. The data used in this study came from two inter-laboratory studies for wheat grain analysis organized in 2016 in the framework of the REQUASUD network. The results showed that improved predictions in terms of prediction errors can be obtained using local approaches compared to the classical global PLS. Moreover, the study highlighted clear differences between inter-laboratory studies and participating laboratories, which were even more evident when working with local procedures.","PeriodicalId":20429,"journal":{"name":"Proceedings of the 18th International Conference on Near Infrared Spectroscopy","volume":"7 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90917880","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
J. A. Adame-Siles, D. Pérez-Marín, J. Guerrero-Ginel, A. Larsen, A. Garrido-Varo
Author Summary: Feed grains are typically transported in bulk and a statistically representative sample of the grain in the truckload is usually required to be taken to the laboratory for wet chemistry or at-line near infrared (NIR) spectroscopy analysis. Currently, most methodologies make use of a physical sampling probe, which mechanically or pneumatically withdraws samples from various depths. Nevertheless, not only is the implementation of this approach expensive and time-consuming, but it is also limited by low sample throughput. In this context, the authors’ group is involved in a large research and development project to find more efficient and cost-effective ways of sampling and analyzing bulk raw materials at the reception level. This work presents a piece of this research focused on the evaluation of the optical performance of two fiber-optic probes designed for automated use as immersion probes in truckloads. It is worth noting the rather different optical design of these two diffuse reflectance probes. Probe A features eight bundles (37 fibers/bundle), four for measurement and four for illumination, 0.5 m in length, and four sapphire windows located around the probe diameter. Probe B has one fiber-optic bundle for measurement (7 fibers) and one for illumination (19 fibers), 3 m in length, and a stainless-steel head with two sapphire windows. The experimental design of this laboratory study aimed at imitating the control of bulk lots of two sort of cereals (maize and wheat). For this purpose, a sample of each cereal was placed into a container (0.34 m in width, 0.4 m in length and 0.25 in height) for analysis. To avoid interferences caused by design, both probes were attached to the same Fourier transform-NIR instrument (Matrix-F, Bruker Optics), and spectra were acquired in the range 834.2–2502.4 nm using the same settings. Two different strategies for recording reference spectra were followed in each case (before the first scan and either after every measurement or after every set of 10 measurements). Noisy regions and spectral repeatability were assessed as a first step towards the evaluation of the feasibility of these probes for performing on-site analysis.
作者总结:饲料谷物通常是散装运输的，通常需要将卡车上具有统计代表性的谷物样本带到实验室进行湿化学或近红外光谱分析。目前，大多数方法使用物理采样探头，通过机械或气动方式从不同深度提取样本。然而，这种方法的实现不仅昂贵且耗时，而且还受到低样本吞吐量的限制。在这种情况下，作者小组参与了一项大型研究和开发项目，以寻找更有效和更具成本效益的方法，在接收层面取样和分析散装原材料。本研究的重点是评估两种光纤探头的光学性能，这两种光纤探头设计用于卡车自动使用的浸入式探头。值得注意的是，这两种漫反射探头的光学设计相当不同。探头A有8束(37根纤维/束)，4束用于测量，4束用于照明，长度为0.5 m，四个蓝宝石窗口位于探头直径周围。探头B有一个用于测量的光纤束(7根纤维)和一个用于照明的光纤束(19根纤维)，长度3米，一个不锈钢头和两个蓝宝石窗口。本实验室研究的实验设计旨在模仿两种谷物(玉米和小麦)的散装控制。为此，将每种谷物的样品放入一个容器(宽0.34米，长0.4米，高0.25米)中进行分析。为了避免设计造成的干扰，两个探针连接到相同的傅里叶变换-近红外仪器(Matrix-F, Bruker Optics)上，使用相同的设置在834.2-2502.4 nm范围内获得光谱。在每种情况下，采用两种不同的记录参考光谱的策略(在第一次扫描之前、每次测量之后或每组10次测量之后)。对噪声区域和光谱重复性进行评估是评估这些探针进行现场分析可行性的第一步。
{"title":"Assessing the potential of two customized fiber-optic probes for on-site analysis of bulk feed grains","authors":"J. A. Adame-Siles, D. Pérez-Marín, J. Guerrero-Ginel, A. Larsen, A. Garrido-Varo","doi":"10.1255/NIR2017.003","DOIUrl":"https://doi.org/10.1255/NIR2017.003","url":null,"abstract":"Author Summary: Feed grains are typically transported in bulk and a statistically representative sample of the grain in the truckload is usually required to be taken to the laboratory for wet chemistry or at-line near infrared (NIR) spectroscopy analysis. Currently, most methodologies make use of a physical sampling probe, which mechanically or pneumatically withdraws samples from various depths. Nevertheless, not only is the implementation of this approach expensive and time-consuming, but it is also limited by low sample throughput. In this context, the authors’ group is involved in a large research and development project to find more efficient and cost-effective ways of sampling and analyzing bulk raw materials at the reception level. This work presents a piece of this research focused on the evaluation of the optical performance of two fiber-optic probes designed for automated use as immersion probes in truckloads. It is worth noting the rather different optical design of these two diffuse reflectance probes. Probe A features eight bundles (37 fibers/bundle), four for measurement and four for illumination, 0.5 m in length, and four sapphire windows located around the probe diameter. Probe B has one fiber-optic bundle for measurement (7 fibers) and one for illumination (19 fibers), 3 m in length, and a stainless-steel head with two sapphire windows. The experimental design of this laboratory study aimed at imitating the control of bulk lots of two sort of cereals (maize and wheat). For this purpose, a sample of each cereal was placed into a container (0.34 m in width, 0.4 m in length and 0.25 in height) for analysis. To avoid interferences caused by design, both probes were attached to the same Fourier transform-NIR instrument (Matrix-F, Bruker Optics), and spectra were acquired in the range 834.2–2502.4 nm using the same settings. Two different strategies for recording reference spectra were followed in each case (before the first scan and either after every measurement or after every set of 10 measurements). Noisy regions and spectral repeatability were assessed as a first step towards the evaluation of the feasibility of these probes for performing on-site analysis.","PeriodicalId":20429,"journal":{"name":"Proceedings of the 18th International Conference on Near Infrared Spectroscopy","volume":"26 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91222287","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
L. Marinoni, A. Stroppa, S. Barzaghi, K. Cremonesi, Nicolò Pricca, A. Meucci, Giulia Pedrolini, Andrea Galli, G. Cabassi
Author Summary: The GRANIR project founded by the Grana Padano Protection Consortium and developed by CREA-ZA research centre is devoted to the development of a rapid and economic method for the chemical characterisation of Grana Padano PDO cheese based on near infrared (NIR) spectroscopy technology. For this purpose, the Consortium purchased several portable spectrometers XNIRTM (dinamica generale®, Poggio Rusco, MN, Italy), to be assigned to the Consortium staff for screening operations of production batches in the fire-branding step, in warehouses and at the packaging step, on cheese paste. To develop predictive models and to evaluate the performance of the portable instruments, 195 samples of Grana Padano were scanned directly on the whole open wheel, scanning both rind and cheese paste. Robust models were built for the prediction of dry matter, fat, fat/dry matter, proteins and proteins/dry matter content using average spectra of rind and paste and chemical data of cheese paste. Additional spectra acquired with two other instruments were included in order to make the models less sensitive to different instruments. Spectra of the same samples acquired at different temperatures (10, 16 and 25 °C) were also added to the dataset in order to reduce the influence of temperature on prediction results. The obtained results showed a satisfactory predictive ability of the models built with portable NIR spectrometers, with respect to the chemical composition of Grana Padano cheese, showing root mean square errors in prediction comparable to that obtained with a Fourier-Transform NIR benchtop instrument. This allows the estimation of average cheese composition, at batch level, using multiple scans taken on a high number of wheels.
{"title":"On site monitoring of Grana Padano cheese production using portable spectrometers","authors":"L. Marinoni, A. Stroppa, S. Barzaghi, K. Cremonesi, Nicolò Pricca, A. Meucci, Giulia Pedrolini, Andrea Galli, G. Cabassi","doi":"10.1255/NIR2017.085","DOIUrl":"https://doi.org/10.1255/NIR2017.085","url":null,"abstract":"Author Summary: The GRANIR project founded by the Grana Padano Protection Consortium and developed by CREA-ZA research centre is devoted to the development of a rapid and economic method for the chemical characterisation of Grana Padano PDO cheese based on near infrared (NIR) spectroscopy technology. For this purpose, the Consortium purchased several portable spectrometers XNIRTM (dinamica generale®, Poggio Rusco, MN, Italy), to be assigned to the Consortium staff for screening operations of production batches in the fire-branding step, in warehouses and at the packaging step, on cheese paste. To develop predictive models and to evaluate the performance of the portable instruments, 195 samples of Grana Padano were scanned directly on the whole open wheel, scanning both rind and cheese paste. Robust models were built for the prediction of dry matter, fat, fat/dry matter, proteins and proteins/dry matter content using average spectra of rind and paste and chemical data of cheese paste. Additional spectra acquired with two other instruments were included in order to make the models less sensitive to different instruments. Spectra of the same samples acquired at different temperatures (10, 16 and 25 °C) were also added to the dataset in order to reduce the influence of temperature on prediction results. The obtained results showed a satisfactory predictive ability of the models built with portable NIR spectrometers, with respect to the chemical composition of Grana Padano cheese, showing root mean square errors in prediction comparable to that obtained with a Fourier-Transform NIR benchtop instrument. This allows the estimation of average cheese composition, at batch level, using multiple scans taken on a high number of wheels.","PeriodicalId":20429,"journal":{"name":"Proceedings of the 18th International Conference on Near Infrared Spectroscopy","volume":"4 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89942589","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
E. Alamu, B. Maziya-Dixon, T. Z. Felde, P. Kulakow, E. Parkes
Author Summary: A developed Near Infrared Reflectance Spectroscopy (NIRS) calibration equation was used for determining provitamin A carotenoids contents of different trials of fresh yellow root cassava genotypes using a total of 50 cassava genotypes scanned twice by NIRS from 400 nm to 2498 nm. The NIRS calibration equations were used to predict the 2-cryptoxanthin, 13-cis β-carotene, trans β-carotene, 9-cis β-carotene, total β-carotene and total carotenoid concentrations of the samples. The predicted values for total carotenoids (TC-pred) ranged from 3.93 μg g–1 to 10.51 μg g–1 with mean of 7.07 ± 2.55 μg g–1 for International Collaborative Trials (ICT), 7.97–11.03 μg g–1 fresh weight with mean of 9.40 ± 0.76 μg g–1 for yellow root trial 8 (Multi-location Uniform Yield Trial) and 6.38–10.44 μg g–1 with mean of 8.74 ± 1.07 μg g–1 for yellow root trial 9 (Multilocation Advanced Yield Trial). Total carotenoids results using reference spectrophotometric method (TC-spec) ranged from 2.57 μg g–1 to 9.97 μg g–1 with mean of 5.66 ± 2.99 μg g–1 for ICT, 6.55–8.74 μg g–1 with mean of 7.74 ± 0.64 μg g–1 for yellow root trial 8 and 4.22–11.00 μg g–1 with mean of 7.57 ± 1.54 μg g–1 for yellow root trail 9. There is significant (P ≤ 0.001) positive correlation (r = 0.55) between TC-pred by NIRS and TC-spec. Also, significant (P ≤ 0.001) positive correlation (r = 0.52) exist between trans β-carotene predicted by NIRS and high-performance liquid chromatography reference. The developed NIRS calibration equations could be used to predict total carotenoids and trans β-carotene content of yellow root cassava and serve as rapid and cost-effective screening method for large cassava sample sets.
{"title":"Application of near infrared reflectance spectroscopy in screening of fresh cassava (Manihot esculenta crantz)\u0000storage roots for provitamin A carotenoids","authors":"E. Alamu, B. Maziya-Dixon, T. Z. Felde, P. Kulakow, E. Parkes","doi":"10.1255/NIR2017.091","DOIUrl":"https://doi.org/10.1255/NIR2017.091","url":null,"abstract":"Author Summary: A developed Near Infrared Reflectance Spectroscopy (NIRS) calibration equation was used for determining provitamin A carotenoids contents of different trials of fresh yellow root cassava genotypes using a total of 50 cassava genotypes scanned twice by NIRS from 400 nm to 2498 nm. The NIRS calibration equations were used to predict the 2-cryptoxanthin, 13-cis β-carotene, trans β-carotene, 9-cis β-carotene, total β-carotene and total carotenoid concentrations of the samples. The predicted values for total carotenoids (TC-pred) ranged from 3.93 μg g–1 to 10.51 μg g–1 with mean of 7.07 ± 2.55 μg g–1 for International Collaborative Trials (ICT), 7.97–11.03 μg g–1 fresh weight with mean of 9.40 ± 0.76 μg g–1 for yellow root trial 8 (Multi-location Uniform Yield Trial) and 6.38–10.44 μg g–1 with mean of 8.74 ± 1.07 μg g–1 for yellow root trial 9 (Multilocation Advanced Yield Trial). Total carotenoids results using reference spectrophotometric method (TC-spec) ranged from 2.57 μg g–1 to 9.97 μg g–1 with mean of 5.66 ± 2.99 μg g–1 for ICT, 6.55–8.74 μg g–1 with mean of 7.74 ± 0.64 μg g–1 for yellow root trial 8 and 4.22–11.00 μg g–1 with mean of 7.57 ± 1.54 μg g–1 for yellow root trail 9. There is significant (P ≤ 0.001) positive correlation (r = 0.55) between TC-pred by NIRS and TC-spec. Also, significant (P ≤ 0.001) positive correlation (r = 0.52) exist between trans β-carotene predicted by NIRS and high-performance liquid chromatography reference. The developed NIRS calibration equations could be used to predict total carotenoids and trans β-carotene content of yellow root cassava and serve as rapid and cost-effective screening method for large cassava sample sets.","PeriodicalId":20429,"journal":{"name":"Proceedings of the 18th International Conference on Near Infrared Spectroscopy","volume":"5 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2019-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81347118","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：