Near-infrared Calibration Models Research Articles

Comparative Analysis of Machine Learning and Deep Learning Algorithms for Assessing Agricultural Product Quality Using NIRS.

The success of near-infrared spectroscopy (NIRS) analysis hinges on the precision and robustness of the calibration model. Shallow learning (SL) algorithms like partial least squares discriminant analysis (PLS-DA) often fall short in capturing the interrelationships between adjacent spectral variables, and the analysis results are easily affected by spectral noise, which dramatically limits the breadth and depth of applications of NIRS. Deep learning (DL) methods, with their capacity to discern intricate features from limited samples, have been progressively integrated into NIRS. In this paper, two discriminant analysis problems, including wheat kernels and Yali pears as examples, and several representative calibration models were used to research the robustness and effectiveness of the model. Additionally, this article proposed a near-infrared calibration model, which was based on the Gramian angular difference field method and coordinate attention convolutional neural networks (G-CACNNs). The research results show that, compared with SL, spectral preprocessing has a smaller impact on the analysis accuracy of consensus learning (CL) and DL, and the latter has the highest analysis accuracy in the modeling results using the original spectrum. The accuracy of G-CACNNs in two discrimination tasks was 98.48% and 99.39%. Finally, this research compared the performance of various models under noise to evaluate the robustness and noise resistance of the proposed method.

Blend uniformity monitoring in a continuous manufacturing mixing process for a low-dosage formulation using a stream sampler and near infrared spectroscopy

Continuous manufacturing has the potential to offer several benefits for the production of oral solid dosage forms, including reduced costs, low-scale equipment, and the application of process analytical technology (PAT) for real-time process control. This study focuses on the implementation of a stream sampler to develop a near infrared (NIR) calibration model for blend uniformity monitoring in a continuous manufacturing mixing process. Feeding and mixing characterizations were performed for three loss-in-weight feeders and a commercial continuous mixer to prepare powder blends of 2.5–7.5 % w/w ibuprofen DC 85 W with a total throughput of 33 kg/h. The NIR spectral acquisition was performed after the mixing stage using a stream sampler for flowing powders. A continuous mixer shaft speed of 250 RPM was selected to operate the mixing process based on a variability analysis developed with in-line spectral data acquired using the stream sampler at 6 RPM. A partial least squares regression (PLS-R) model was performed and evaluated, yielding a root-mean-square error of prediction (RMSEP) of 0.39 % w/w and a bias of 0.05 % w/w. An independent experimental run conducted two days later revealed that the continuous mixing process and the NIR calibration model presented low day-to-day variation. The minimum practical error (MPE) and sill values through variographic analysis showed low variance associated with the sampling process using the stream sampler. Results demonstrated the promising capacity of the stream sampler coupled to an NIR probe to be implemented within continuous manufacturing processes for the real-time determination of API concentration.

Use of Constant-Analyte Samples in Updating Quantitative Near-Infrared Calibration Models for Physiological Levels of Glucose

Updating procedures were investigated for multivariate calibration models for physiological levels of glucose in aqueous samples based on near-infrared spectroscopy. The methodology evaluated used samples with constant glucose concentrations collected on the prediction day to augment previously collected calibration spectra for the purpose of computing an updated model for use only on that day. Models were based on partial-least squares (PLS) regression in conjunction with spectral preprocessing using direct orthogonal signal correction (DOSC). The use of constant-analyte samples allowed the procedure to be compatible with samples collected at the beginning of a chemical reaction or from a fasting patient in a blood glucose monitoring application. The methodology was tested with nine sets of prediction spectra collected from 368 to 617 days after the collection of the calibration data. The updated PLS models based on four augmented samples exhibited an approximately 50% improvement in prediction accuracy relative to models that were based solely on the original calibration data. The use of DOSC in conjunction with updating was found to be beneficial overall, but only led on average to less than 10% improvement in prediction performance.

New approach for near-infrared wavelength selection using a combination of MIC and firefly evolution

Full-length spectral data analysis has a big problem that the variables are highly in collinearity and correlation. Spectral wavelength selection is a continuing hot topic in quantitative or qualitative analysis. In this paper, we propose a new approach for near-infrared (NIR) wavelength selection. The novel strategy mainly refers to the modification of maximum information coefficient (MIC) method and an improvement of firefly evolutionary algorithm. We introduce the orthogonal decomposition to modify the MIC method, so as to search the informative signals conceived in projection vectors. We also raise the common firefly algorithm (FA) as in the discretized mode, and design a novel adaptive mapping function to improve its intelligent computing effect. In experiment, the modified MIC (MICm) method and the adaptive discrete FA algorithm (DFAadp) are joint together for combined optimization of the NIR calibration model. The proposed combined modeling strategy is applied for quantitative analysis of the fishmeal samples, in the concern to select their informative variables/wavelengths. Experimental results indicate that the combination of MICm and DFAadp perform better than traditional MIC method and common DFA. We conclude that the proposed combined optimization strategy is beneficial for wavelength selection in NIR spectral analysis. It is anticipated to be validated for further applications in a wide range.

Near-infrared spectroscopy (NIRS) for monitoring the nutritional composition of black soldier fly larvae (BSFL) and frass.

The demand for protein obtained from animal sources is growing rapidly, as is the necessity for sustainable animal feeds. The use of black soldier fly larvae (BSFL) reared on organic side streams as sustainable animal feed has been receiving attention lately. This study assessed the ability of near-infrared spectroscopy (NIRS) combined with chemometrics to evaluate the nutritional profile of BSFL instars (fifth and sixth) and frass obtained from two different diets, namely soy waste and customised bread-vegetable diet. Partial least squares (PLS) regression with leave one out cross-validation was used to develop models between the NIR spectral data and the reference analytical methods. Calibration models with good [coefficient of determination in calibration (Rcal 2 ): 0.90; ratio of performance to deviation (RPD) value: 3.6] and moderate (Rcal 2 : 0.76; RPD value: 2.1) prediction accuracy was observed for acid detergent fibre (ADF) and total carbon (TC), respectively. However, calibration models with moderate accuracy were observed for the prediction of crude protein (CP) (Rcal 2 : 0.63; RPD value: 1.4), crude fat (CF) (Rcal 2 : 0.70; RPD value: 1.6), neutral detergent fibre (NDF) (Rcal 2 : 0.60; RPD value: 1.6), starch (Rcal 2 : 0.52; RPD value: 1.4), and sugars (Rcal 2 : 0.52; RPD value: 1.4) owing to the narrow or uneven distribution of data over the range evaluated. The near-infrared (NIR) calibration models showed a good to moderate prediction accuracy for the prediction of ADF and TC content for two different BSFL instars and frass reared on two different diets. However, calibration models developed for predicting CP, CF, starch, sugars and NDF resulted in models with limited prediction accuracy. © 2023 Society of Chemical Industry.

Improving transferability and service life of the calibration model of total plant alkaloids in tobacco leaves on seven NIR spectroscopy devices by multi-step wavelength selection methods

The rapid determination of total plant alkaloids (TPA) content in tobacco products by means of near infrared (NIR) spectroscopy is becoming conventional detection method in tobacco industry. An established NIR model of TPA is anticipated to be shared by other devices as long as possible not needing to update the model. Present study developed a three-step wavelength selection method SISCW (selecting important and stable characteristic wavelengths）to build a robust NIR calibration model of TPA to improve its transferability and to prolong its use period at many devices. For this, 292 flue-cured tobacco leaf samples from more than 10 regions of China, which were harvested in 2011–2013 and tested on the primary NIR device, were used to build the model. 77 samples harvested from 2011 to 2013 were applied to verify the transferability of the model on six secondary NIR devices. 180 samples harvested from 2014 to 2020 that were divided into 7 groups according to their growing years, were used to examine the long-term application capacity of the model on the seven NIR devices. The SISCW method integrates an image processing method of scale invariant feature transformation (SIFT) with analyzing standard deviation of the sample spectra (SDSS) and water absorption coefficients to select important and stable characteristic wavelengths (SISCW). The wavelengths selected by the SISCW were recorded as Uisc. The results indicated that the partial least square (PLS) calibration model of TPA based on Uisc (SISCW-PLS) performed well on both primary and the six secondary devices when it was employed to predict TPA of the 77 samples of 2011–2013. The model has run 7 years on the primary and 3 of the secondary devices from 2014 to 2020. This emphasizes the SISCW method has great potential for improving transferability and service-life of TPA calibration model.

Characterizing Spray-Dried Powders through NIR Spectroscopy: Effect of Two Preparation Strategies for Calibration Samples and Comparison of Two Types of NIR Spectrometers.

Emerging portable near infrared (NIR) spectroscopic approaches coupled with data analysis and chemometric techniques provide opportunities for the rapid characterization of spray-dried products and process optimization. This study aimed to enhance the understanding of applying NIR spectroscopy in spray-dried samples by comparing two sample preparation strategies and two spectrometers. Two sets of whey protein-maltodextrin matrixes, one with a protein content gradient and one with a consistent protein content, were spray-dried, and the effect of the two preparation strategies on NIR calibration model development was studied. Secondly, a portable NIR spectrometer (PEAK) was compared with a benchtop NIR spectrometer (CARY) for the moisture analysis of prepared samples. When validating models with the samples with focused protein contents, the best PLS protein models established from the two sample sets had similar performances. When comparing two spectrometers, although CARY outperformed PEAK, PEAK still demonstrated reliable performance for moisture analysis, indicating that it is capable as an inline sensor.

Model averaging in calibration of near-infrared instruments with correlated high-dimensional data

Model averaging (MA) is a modelling strategy where the uncertainty in the configuration of selected variables is taken into account by weight-combining each estimate of the so-called ‘candidate model’. Some studies have shown that MA enables better prediction, even in high-dimensional cases. However, little is known about the model prediction performance at different types of multicollinearity in high-dimensional data. Motivated by calibration of near-infrared (NIR) instruments,we focus on MA prediction performance in such data. The weighting schemes that we consider are based on the Akaike’s information criterion (AIC), Mallows’ Cp , and cross-validation. For estimating the model parameters, we consider the standard least squares and the ridge regression methods. The results indicate that MA outperforms model selection methods such as LASSO and SCAD in high-correlation data. The use of Mallows’ Cp and cross-validation for the weights tends to yield similar results in all structures of correlation, although the former is generally preferred. We also find that the ridge model averaging outperforms the least-squares model averaging. This research suggests ridge model averaging to build a relatively better prediction of the NIR calibration model.

Classification of waste cotton from different countries using the near-infrared technique

Cotton fiber is mainly composed of cellulose, which is rarely separated into different kinds. However, the classification and identification of waste cotton from different countries are essential for the customs service of the country. In this study, the near-infrared classification method was introduced to classify and identify cotton fibers. Waste cotton samples from six different countries were collected, and one-fifth of them were used for validation. The near-infrared calibration and prediction models were constructed using both soft independent modeling of class analogy and partial least squares methods. It was found that the optimized model has a high recognition rate, and the prediction accuracy of the model was 99% for six countries. It was demonstrated that the near-infrared model established in this study can be used for fast and accurate identification of waste cotton from different countries.

Estimation of crude protein and amino acid contents in whole, ground and defatted ground soybeans by different types of near-infrared (NIR) reflectance spectroscopy

This study was designed to examine the ability of near-infrared (NIR) spectroscopy to determine amino acid and crude protein concentrations in soybean and explore the effects of particle size, fat content and types of NIR spectroscopy instruments on the predictive ability of the NIR models. Whole, ground and defatted ground soybean were scanned by a diode array NIR analyzer and defatted ground soybean were scanned by a Fourier transformed NIR analyzer. Samples were analyzed for crude protein and amino acid contents using reference wet chemistry methods. The NIR calibration models of crude protein and most amino acids showed acceptable coefficient of determinations (R2c = 0.605–0.952), while the NIR models for Cys, Met and Trp were less accurate (R2c = 0.498–0.667 for Cys and R2c = 0.482–0.615 for Met and R2c = 0.406–0.481 for Trp). The R2c values in NIR calibration were found to relate to the correlation between amino acids and crude protein. The grinding process and lipid extraction were found to improve the NIR spectroscopy in crude protein and amino acids prediction. The Fourier transformed NIR analyzer was found to perform similar to the diode array NIR analyzer for defatted ground soybean analysis. No significant differences were found between crude protein and amino acid contents predicted by these two NIR spectroscopy instruments (P > 0.05).

Evaluation of Near-Infrared Reflectance and Transflectance Sensing System for Predicting Manure Nutrients

Livestock manure is widely applied onto agriculture soil to fertilize crops and increase soil fertility. However, it is difficult to provide real-time manure nutrient data based on traditional lab analyses during application. Manure sensing using near-infrared (NIR) spectroscopy is an innovative, rapid, and cost-effective technique for inline analysis of animal manure. This study investigated a NIR sensing system with reflectance and transflectance modes to predict N speciation in dairy cow manure using a spiking method. In this study, 20 dairy cow manure samples were collected and spiked to achieve four levels of ammoniacal nitrogen (NH4-N) and organic nitrogen (Org-N) concentrations that resulted in 100 samples in each spiking group. All samples were scanned and analyzed using a NIR system with reflectance and transflectance sensor configurations. NIR calibration models were developed using partial least square regression analysis for NH4-N, Org-N, total solid (TS), ash, and particle size (PS). Coefficient of determination (R2) and root mean square error (RMSE) were selected to evaluate the models. A transflectance probe with a 1 mm path length had the best performance for analyzing manure constituents among three path lengths. Reflectance mode improved the calibration accuracy for NH4-N and Org-N, whereas transflectance mode improved the model predictability for TS, ash, and PS. Reflectance provided good prediction for NH4-N (R2 = 0.83; RMSE = 0.65 mg mL−1) and approximate predictions for Org-N (R2 = 0.66; RMSE = 1.18 mg mL−1). Transflectance was excellent for TS predictions (R2 = 0.97), and provided good quantitative predictions for ash and approximate predictions for PS. The correlations between the accuracy of NH4-N and Org-N calibration models and other manure parameters were not observed indicating the predictions of N contents were not affected by TS, ash, and PS.

Machine Learning Framework for Intelligent Detection of Wastewater Pollution by IoT-Based Spectral Technology

Industrial wastewater contains excessive micro insoluble solids (MIS) that probably cause environmental pollutions. Near-infrared (NIR) spectroscopy is an advanced technology for rapid detection of the complex targets in wastewater. An Internet of Things (IoT) platform would support intelligent application of the NIR technologies. The studies of intelligent chemometric methods mainly contribute to improve the NIR calibration model based on the IoT platform. With the development of artificial intelligence, the backward interval and synergy interval techniques were proposed in combination use with the least square support vector machine (LSSVM) method, for adaptive selection of the informative spectral wavelength variables. The radial basis function (RBF) kernel is applied for nonlinear mapping. The regulation parameter and the kernel width are fused together for smart optimization. In the design for waveband autofittings, the total of digital wavelengths in the full scanning range was split into 43 equivalent subintervals, and then, the back interval LSSVM (biLSSVM) and the synergy interval LSSVM (siLSSVM) models were both established for the improvement of prediction results based on the adaptive selection of quasidiscrete variable combination. In comparison with some common linear and nonlinear models, the best training model was acquired with the siLSSVM method while the best testing model was obtained with biLSSVM. The intelligent optimization of model parameters indicated that the proposed biLSSVM and siLSSVM deep learning methodologies are feasible to improve the model prediction results in rapid determination of the wastewater MIS content by the IoT-based NIR technology. The machine learning framework is prospectively applied to the fast assessment of the environmental risk of industrial pollutions and water safety.

Rapid and comprehensive quality assessment of Bupleuri Radix through near-infrared spectroscopy combined with chemometrics

The quality of popular Chinese herbal medicine Bupleuri Radix (BR) is essential to its curative effect and economic value and has become an issue for customers and dealers. Traditional analytical techniques for evaluating BR quality are generally time-consuming, laborious, and costly. In this regard, a rapid and efficient method is warranted to supplement existing approaches. This study developed a comprehensive strategy based on near-infrared (NIR) spectroscopy and chemometrics for quality assessment of BR. From qualitative perspective, BR was identified according to geographical origin by using principal component analysis (PCA), partial least square-discriminant analysis (PLS-DA), and support vector machine (SVM). Saikosaponin A, saikosaponin D, polysaccharides, and total flavonoids, which are vital quality indicators of BR, were quantified through NIR spectroscopy and calibration models. Herein, a CARS-SVM model was constructed based on competitive adaptive reweighted sampling and SVM algorithm, and its prediction performance was compared with other calibration models (Full-PLS, CARS-PLS, and Full-SVM). Results indicated that the SVM model performed better than PCA and PLS-DA and had correct recognition rates of 97.37% and 94.44% in the calibration and prediction sets, respectively. The contents of saikosaponin A, saikosaponin D, polysaccharides, and total flavonoids were satisfactorily predicted through the CARS-SVM model. The coefficient of determination (Rp2) and root mean square error in the independent test sets (RMSEP) were 0.9135 and 0.0837%, 0.8132 and 0.1695%, 0.8810 and 0.7165%, and 0.7471 and 0.1766% for saikosaponin A, saikosaponin D, polysaccharides, and total flavonoids, respectively. The superiority of CARS-SVM compared with the other quantitative models demonstrated that CARS wavelength selection method and nonlinear SVM regression tool significantly improved the model performance. Overall the results demonstrated the potential of NIR spectroscopy and appropriate chemometrics for rapid quality evaluation of BR. The developed comprehensive quality assessment strategy could supplement existing methods for analysis of BR and is expected to provide reference for quality evaluation and control of other Chinese herbal medicines.

Rapid Measurement of Cellulose, Hemicellulose, and Lignin Content in Sargassum horneri by Near-Infrared Spectroscopy and Characteristic Variables Selection Methods.

Near-infrared (NIR) spectroscopy and characteristic variables selection methods were used to develop a quick method for the determination of cellulose, hemicellulose, and lignin contents in Sargassum horneri. Calibration models for cellulose, hemicellulose, and lignin in Sargassum horneri were established using partial least square regression methods with full variables (full-PLSR). The PLSR calibration models were established by four characteristic variables selection methods, including interval partial least square (iPLS), competitive adaptive reweighted sampling (CARS), correlation coefficient (CC), and genetic algorithm (GA). The results showed that the performance of the four calibration models, namely iPLS-PLSR, CARS-PLSR, CC-PLSR, and GA-PLSR, was better than the full-PLSR calibration model. The iPLS method was best in the performance of the models. For iPLS-PLSR, the determination coefficient (R2), root mean square error (RMSE), and residual predictive deviation (RPD) of the prediction set were as follows: 0.8955, 0.8232%, and 3.0934 for cellulose, 0.8669, 0.4697%, and 2.7406 for hemicellulose, and 0.7307, 0.7533%, and 1.9272 for lignin, respectively. These findings indicate that the NIR calibration models can be used to predict cellulose, hemicellulose, and lignin contents in Sargassum horneri quickly and accurately.

A NIR-Based Study of Desorption Kinetics during Continuous Spin Freeze-Drying

The pharmaceutical industry is progressing toward the development of more continuous manufacturing techniques. At the same time, the industry is striving toward more process understanding and improved process control, which requires the implementation of process analytical technology tools (PAT). For the purpose of drying biopharmaceuticals, a continuous spin freeze-drying technology for unit doses was developed, which is based on creating thin layers of product by spinning the solution during the freezing step. Drying is performed under vacuum using infrared heaters to provide energy for the sublimation process. This approach reduces drying times by more than 90% compared to conventional batch freeze-drying. In this work, a new methodology is presented using near-infrared (NIR) spectroscopy to study the desorption kinetics during the secondary drying step of the continuous spin freeze-drying process. An inline PLS-based NIR calibration model to predict the residual moisture content of a standard formulation (i.e., 10% sucrose) was constructed and validated. This model was then used to evaluate the effect of different process parameters on the desorption rate. Product temperature, which was controlled by a PID feedback mechanism of the IR heaters, had the highest positive impact on the drying rate during secondary drying. Using a higher cooling rate during spin freezing was found to significantly increase the desorption rate as well. A higher filling volume had a smaller negative effect on the drying rate while the chamber pressure during drying was found to have no significant effect in the range between 10 and 30 Pa.

Near-infrared calibration models for estimating volatile fatty acids and methane production from in vitro rumen fermentation of different total mixed rations

Volatile fatty acids (VFA) and methane (CH4) are the major products of rumen fermentation. The VFA are considered an energy source for the animal and rumen microbiota, and CH4 (which is released by eructation) is considered an energy loss. Quantification of these fermentation products is fundamental for the evaluation of feeds and diets, and provides important information regarding the use of nutrients by ruminants. Near-infrared (NIR) spectroscopy is increasingly used for the evaluation of animal feeds because it is rapid, nondestructive, noninvasive, and inexpensive; does not require reagents; and the results are reproducible. The aim of this study was to develop NIR calibration models for estimating the production of VFA (acetic, propionic, butyric, valeric, isovaleric, and isobutyric acids), total gas, and CH4 using in vitro gas production tests with buffered rumen inoculum throughout fermentation. Fifty-four total mixed rations (TMRs) were examined, and rumen fluid was manually collected from 2 dry Holstein dairy cows that had ruminal fistulas and were fed at maintenance energy levels. Then, 30 mL of buffered rumen fluid was incubated in bottles with ~220 mg of TMR. The total gas, VFA, and CH4 were measured after 2, 5, 9, 24, 30, 48, and 72 h of rumen incubation for each TMR. The VFA were measured on 32 randomly selected TMR. In particular, 7 bottles were used for each TMR, one for each incubation time. Methane was measured in the headspace and VFA were measured in the buffered rumen fluid. The bottles were considered experimental units for calibration purposes. The production of CH4 was quantified from the bottle headspaces by gas chromatography, and total gas production was measured using a pressure transducer at each incubation time. Two aliquots of the fermented liquids were sampled by opening the bottles at each incubation time, and (1) the concentrations of VFA were determined by gas chromatography or (2) spectra were obtained from Fourier-transform NIR spectroscopy. The data were randomly divided into calibration and validation data sets. The average concentrations of acetic acid (45.30 ± 11.92 and 43.86 ± 11.93 mmol/L), propionic acid (14.97 ± 6.08 and 14.38 ± 6.56 mmol/L), butyric acid (8.47 ± 3.47 and 8.65 ± 3.79 mmol/L), total gas (111.34 ± 81.90 and 116.46 ± 82.44 mL/g of organic matter), and CH4 (9.65 ± 9.45 and 10.35 ± 9.33 mmol/L) were similar in the 2 data sets. The best calibration models were retained based on the coefficient of determination (R2) and the ratio of prediction to deviation (RPD). The R2 values for prediction of VFA ranged from 0.69 (RPD = 3.28) for valeric acid to 0.94 (RPD = 4.20) for acetic acid. The models also provided good predictions of CH4 (R2 = 0.89, RPD = 3.05) and cumulative gas production (R2 = 0.91, RPD = 3.30). The models described here precisely and accurately estimated the production of CH4 and VFA during in vitro rumen fermentation tests. Validations at additional laboratories may provide more robust calibrations.

Development of Attenuated Total Reflectance Mid-Infrared (ATR-MIR) and Near-Infrared (NIR) Spectroscopy for the Determination of Resistant Starch Content in Wheat Grains.

The chemical method for the determination of the resistant starch (RS) content in grains is time-consuming and labor intensive. Near-infrared (NIR) and attenuated total reflectance mid-infrared (ATR-MIR) spectroscopy are rapid and nondestructive analytical techniques for determining grain quality. This study was the first report to establish and compare these two spectroscopic techniques for determining the RS content in wheat grains. Calibration models with four preprocessing techniques based on the partial least squares (PLS) algorithm were built. In the NIR technique, the mean normalization + Savitzky–Golay smoothing (MN + SGS) preprocessing technique had a higher coefficient of determination (Rc2 = 0.672; Rp2 = 0.552) and a relative lower root mean square error value (RMSEC = 0.385; RMSEP = 0.459). In the ATR-MIR technique, the baseline preprocessing method exhibited a better performance regarding to the values of coefficient of determination (Rc2 = 0.927; Rp2 = 0.828) and mean square error value (RMSEC = 0.153; RMSEP = 0.284). The validation of the developed best NIR and ATR-MIR calibration models showed that the ATR-MIR best calibration model has a better RS prediction ability than the NIR best calibration model. Two high grain RS content wheat mutants were screened out by the ATR-MIR best calibration model from the wheat mutant library. There was no significant difference between the predicted values and chemical measured values in the two high RS content mutants. It proved that the ATR-MIR model can be a perfect substitute in RS measuring. All the results indicated that the ATR-MIR spectroscopy with improved screening efficiency can be used as a fast, rapid, and nondestructive method in high grain RS content wheat breeding.

A brief note on application of domain-invariant PLS for adapting near-infrared spectroscopy calibrations between different physical forms of samples

Near-infrared (NIR) calibration models are widely developed and routinely used for the prediction of physicochemical properties of samples. However, the main challenge with NIR models is that they are highly specific to the physical form of the samples. For example, a NIR calibration established for solid samples can usually not be used for the same samples in powdered form. Domain adaption (DA) techniques, such as domain invariant partial least-squares (di-PLS) regression, have recently appeared in the chemometric domain which allow adapting NIR calibrations for new sample-/instrument- or environment-associated conditions in a standard free manner. A practical use case of di-PLS can be assumed as the adaption of NIR calibration models to be used in different physical forms of samples. In this contribution we show, for the first time, application of di-PLS regression analysis for adapting a near-infrared (NIR) calibration for solid rice kernels to be used on powdered rice flour without the need for new reference measurements for the latter. di-PLS is a domain adaption technique that removes the differences between different but related data sources (i.e. domains) to reach generalized models. The study found that di-PLS allowed a direct adaption of calibration based on solid rice kernels to be used on powdered rice flour without requiring any reference protein measurements for the latter. Our results suggest that DA tools, such as di-PLS, can support a wider usage of chemometric calibrations especially when models need to be adapted to different physical forms of the same samples.

Semi-Supervised Learning-Based Calibration Model Building of NIR Spectroscopy for In Situ Measurement of Biochemical Processes Under Insufficiently and Inaccurately Labeled Samples

To deal with the typical problem of inaccurately and insufficiently labeled samples involved with real measurement of biochemical process components, a semi-supervised learning-based calibration model building method is proposed to use the up-to-date near-infrared (NIR) spectroscopy for in situ measurement of such a process. First, a supervised ladder network is designed to delete the inaccurately labeled samples, such that a cleaned training set is used for establishing the NIR calibration model. To address the issue of insufficiently labeled samples for modeling with the aim of online measurement, a semi-supervised ladder network is explored to establish a regression learning algorithm, by representing the process characteristics from all the unlabeled samples in combination with the cleaned labeled samples. In the ladder network, encoding is used to extract the features and decoding is used for denoising. Meanwhile, the skip connections allow higher layers to concentrate on those features related to the regression task, therefore facilitating the model accuracy. A numerical example and experiments on a glucose fermentation process are shown to demonstrate the effectiveness and advantage of the proposed method.

Real-time quantification of low-dose cohesive formulations within a sampling interface for flowing powders.

This study investigates the performance of a sampling interface for monitoring cohesive, flowing powder formulations with Hausner's Ratio and Carr's Index higher than 1.5 and 35%, respectively. The sampler device was operated in combination with near-infrared (NIR) spectroscopy to quantify ibuprofen concentrations between 1.5 and 4.5% w/w. NIR spectra also provided essential information to study the process dynamics within the sampler. The 200 spectra per blend obtained demonstrated a continuous powder flow with no evidence of agglomerates or segregation within the sampler for a blend of 6kg. A NIR calibration model was optimized to predict independent test blends, delivering root mean square error of predictions and bias under 0.1% w/w. The test blends were within specifications according to the requirements of European Pharmacopeia. Variographic analysis demonstrated that the sampler device may determine low drug concentration in cohesive powder blends, presenting sampling errors below 0.011 (%w/w)2. This analysis also demonstrated that an increase in the blend compressibility leads to a slight rise in sampling errors within the sampler device. The sampler device offers statistical robustness in the evaluation of blend uniformity, providing greater confidence in the quality determination of the cohesive powder blends without significantly affecting its flow properties.