Sense the Real Change: Proceedings of the 20th International Conference on Near Infrared Spectroscopy

This review consists of two parts. The first part is concerned with the outline of the spectra analysis methods in NIR spectroscopy. In this part conventional spectra analysis methods such as the analysis based on group frequencies and calculations of difference spectra and second derivatives are explained. The second part describes big efforts of the Ozaki group in unraveling complicated NIR spectra for last three decades. Our studies using two-dimensional correlation spectroscopy (2D-COS) are mentioned first. The analysis of the temperature-dependent spectra variations of water using difference spectra, 2D-COS, principal component analysis (PCA), and self-modeling curve resolution (SMCR) is reported next. Our proposals of new chemometrics algorisms and 2D-COS algorisms are introduced. Moreover, our recent investigations of quantum chemical calculation studies of NIR spectra are discussed.

Spectrometers, especially those operating in the near-infrared and visible, are today so small and such low cost that they can be embedded in consumer goods or sold directly to the public. This paper outlines what is available today in portable NIR spectroscopy, and how these instruments can be categorized; emerging applications of miniature spectrometers, especially those in consumer goods using embedded spectrometers and marketed directly to consumers; and caveats on these direct-to-consumer instruments.

Quantum mechanical calculations are routinely used as a major support in mid-infrared (MIR) and Raman spectroscopy. In contrast, practical limitations for long time formed a barrier to developing a similar synergy between near-infrared (NIR) spectroscopy with computational chemistry. Recent advances opened the pathway to modeling NIR spectra of various molecules reaching the size of long-chain fatty acids. Theoretical NIR spectra unveil staggering number of overlapping vibrational bands that create complex spectral patterns in many cases hindering our comprehension of NIR spectra but also revealing rich information in NIR spectra.This review article summarizes the most recent accomplishments in the emerging field. Simulation studies of NIR spectra of a variety of compounds are discussed, with particular attention given to those that are significant from the point of view of physical and analytical chemistry. The examples range from basic molecules (alcohols, nitriles, carboxylic acids) through complex molecules with importance to biophysical science (fatty acids, nucleobases) and analytical chemistry (natural drugs, polyphenols, food adulterants) to organic polymers including a set of standard materials used in various industries. Innovation into analytical routines includes e.g. interpreting the chemical information in the performance profiles of miniaturized spectrometers, or explaining the matrix effects in NIR spectra. These advances lead to intelligent design of the analysis with proper comprehension of chemical information. This review article is based on the topic of an invited oral talk presented at NIR-2021 Conference that took place on 18th–21st October 2021 as an online event organized and coordinated from Beijing.

Near-infrared (NIR) spectroscopy has been a powerful technique for both qualitative and quantitative analysis. Due to the highly overlapping of the spectral bands, however, it is difficult to extract structural and quantitative information from the spectral data. Therefore, chemometric methods have been widely applied to enhance the spectral resolution or extract the spectral information, including modeling techniques, spectral preprocessing, variable selection, outlier detection, modeling transfer, etc. These methods provided colorful approaches for improving the models in both quantitative and discrimination analysis, greatly enhanced the applicability of NIR spectroscopy. On the other hand, temperature-dependent near-infrared spectroscopy was developed for analyzing liquid mixtures or aqueous systems. Chemometric methods were also established to build the quantitative models and extract the temperature-induced spectral variations. The former provided powerful tools for predicting the temperature or the concentration of the components, and the latter provided efficient approaches for understanding the structures and the interactions in chemical and biological samples or processes.

Ongoing miniaturization of spectrometers creates a perfect synergy with the common advantages of NIR spectroscopy. The combination of portability and direct on-site application with high-throughput and non-invasive way of analysis is a decisive advantage in various industries. The characteristics of miniaturized NIR sensors are noted versus benchtop spectrometers regarding the performance, applicability, and optimization of methodology. These devices remarkably increase the flexibility of analysis; however, various factors affect their performance in different analytical scenarios. Currently, it is a focused research direction to perform systematical evaluation studies of the accuracy and reliability of various miniaturized spectrometers that are based on different technologies, e.g. Fourier transform (FT)-NIR, micro-optoelectro-mechanical system (MOEMS) Hadamard mask or linear variable filter (LVF) coupled with array detector.Progressing technology is accompanied by innovative data-analytical methods, integrated into the package of a micro-NIR analytical method to improve its accuracy, reliability and applicability. Advanced calibration methods (e.g. artificial neural networks ANN, nonlinear regression) directly improve the performance of miniaturized instruments in challenging analysis and balance the accuracy of these instruments towards laboratory spectrometers. Two-dimensional correlation spectroscopy (2D-COS) provides insight into the relative sensitivity observed between different instruments in specific NIR bands. The quantum mechanical simulation of NIR spectra reveals the wavenumber regions where the best correlated spectral information resides, and unveils the interactions of the target analyte with surrounding matrix, to ultimately enhance the information gathered from the NIR spectra. This set of methods enables the intelligent design of future micro-NIR analyzes, critically important for samples with a complex matrix.

The research and application of near infrared spectroscopy in China has a history of 40 years. This paper mainly introduces the research and application status of near infrared spectroscopy in China, so that the world peers can fully and systematically understand the actual situation of near infrared spectroscopy in China and sense the changes of China from NIR side.

Ginger is well known for its unique pungent flavour and numerous health benefits, both of which can be attributed to the gingerol derivatives. Quantification of these compounds is important for assessing ginger quality for processing purposes, as their levels can vary with cultivar and growing conditions. However, current analytical techniques are too expensive and time-consuming for in-house monitoring of ginger quality. Hence the objective of this work was to investigate the potential for estimating the content of gingerol derivatives in dried, ground ginger using near-infrared spectroscopy. These compounds are found at relatively low concentrations (1000–6000 mg kg−1), which could make detection by NIRS challenging. In this study, reflectance NIR spectra between 10,000–4,000 cm−1 were collected from 100 dried, powdered samples, using a spinning cup and integrating sphere module on an Thermo Scientific Antaris II FT-NIR Analyzer. The 6-gingerol, 8-gingerol, 10-gingerol and 6-shogaol content of the samples was measured in 90% methanol extracts using high-performance liquid chromatography (HPLC). The performance of partial least squares regression (PLSR) and support vector regression (SVR) were then evaluated, with results validated against a dependent test set. The NIRS method showed promise for the prediction of 6-gingerol content, with $${\text{R}}_{{{\text{cv}}}}^{2}$$ R cv 2  > 0.7, RMSECV < 450 mg kg−1 (on a mean concentration of 4400 mg kg−1) and RPD of approximately 2. Similarly, prediction of 6-shogaol content gave an $${\text{R}}_{{{\text{cv}}}}^{2}$$ R cv 2  > 0.6, RMSECV < 110 mg kg−1 (on a mean concentration of 1440 mg kg−1) and RPD of 1.6. A higher accuracy was found for PLSR compared to SVR. Although the results could be further improved, the detection of these compounds at low concentrations in a matrix as complex as ginger is notable. With further refinement, NIRS may be suitable for the rapid estimation of major pungent compounds in dried ginger.

Quick and easy spectral discrimination technology of milk powder adulterated has important application prospects. The separation degree spectrum between the two spectral populations and a wavelength selection method based on the separation degree priority combination (SDPC) were proposed here. The wavelength step by step phase-out (WSP) method was used for secondary wavelength optimization. Using Vis-NIR spectroscopy combined with SDPC-WSP-PLS-DA method, the discriminant analysis models for milk powder adulterated were established. Among them, standard normal transformation (SNV) and Norris derivative filtering (NDF) were used in turn for the spectral preprocessing with the NDF parameters of d = 2, s = 11, g = 5. The selected optimal SDPC-WSP-PLS-DA model used only 8 wavelengths to achieve high-precision discrimination effect in modeling (100%) and independent validation (96.0%). The results shower the feasibility of applying Vis-NIR spectroscopy to high-precision discriminant analysis of milk powder adulteration. The proposed spectral separation degree method can enhance the spectral difference of different spectral populations, extract information wavelengths, and improve the discrimination effect.

To effectively classify imported frozen fish, we propose a characteristic wavelength selection method based on two-dimensional correlation spectroscopy (2DCOS), which reduces spectral variables required for analysis and improves the accuracy and efficiency of classification, among the data obtained by near-infrared spectroscopy (NIRS). In the experiments, near-infrared spectral were collected from Pollachius, Theragra chalcogramma, Gadous macrocephaius, and Melanogrammus aeglefinus of the family Gadidae, comparing different preprocessing algorithms and selecting multiple scattering corrections. The 2DCOS between the four Cod samples were then constructed. Based on the autocorrelation spectrum of the synchronous 2DCOS, the relative intensities at wavelengths 1580 nm, 1744 nm, and 1900 nm were obtained to be almost zero, and the two highest peaks in the autocorrelation spectrum were at 1550–1580 nm and 1744–1900 nm, as well as the spectra in these two bands, were highly correlated, so the two bands 1550–1580 nm and 1744–1900 nm were filtered out from the complete spectrum. The results are the accuracy of the training set of the waveband SVM filtered based on the 2DCOS is 94.58%, and the accuracy of the validation set can reach up to 97.30%. The study shows that the proposed spectral data compression method based on the 2DCOS technique has a high compression rate and high classification accuracy.

A handheld near infrared (NIR) spectroscopy device, with a wavelength range from 900 nm to 1650 nm and coupled with two Partial Least-Squares Discriminant Analysis (PLS-DA) models, has been used to demonstrate its applicability as a proof of concept for quality monitoring of bigeye tuna (Thunnus obesus). First, a classification model was created to discriminate between injected and non-injected tuna samples. Then, a second classification model was developed to discriminate between non-injected and each water and additives treatment used. The results were promising, showing both models good results in the validation dataset. The first model, with 8 latent variables (LV), had an error-rate of 0.08 and an accuracy value of 0.93. It showed a good discrimination between injected and non-injected samples. The second model, with 10 LV, presented an error rate of 0.15 and an accuracy of 0.88. The discrimination between treatments was good even when protein hydrolysate solutions were used (sensitivity = 0.81; specificity = 0.99; precision = 0.87), a case which is typically hard to detect with accurate destructive analysis. This work opens new possibilities for onsite inspection in the fish industry, where NIR could be used as a complementary tool for the detection of water solutions in tuna.

In this research, NIR combined with chemometrics is applied to identification of abalone variety and age in order to decrease losses in sales and farming. Identification of Green disc abalone age can be realized using principal component analysis (PCA). Partial least square discrimination analysis (PLS-DA) can be divided into two categories: PLS2-DA and PLS1-DA. When a dataset has only two classes, performances of PLS2-DA and the novel approach Euclidean distance coupled to PLS1-DA (EuD-PLS1-DA) are highly similar with accuracy being over 98% for identification of Nan-Ri abalone age and identification of Green disc abalone age. EuD-PLS1-DA is superior to PLS2-DA when confronting multi-class problems, such as variety classification of abalones. Accuracy of EuD-PLS1-DA is 86.52% and 93.38% separately for calibration set and validation set, which is satisfactory; accuracy of PLS2-DA is smaller than 80%. The classification results show the usefulness of NIR linked to PLS-DA for identification of variety and age of abalones.

A discriminant method for adulterated milk was proposed using temperature-dependent two-dimensional (2D) near-infrared (NIR) correlation spectroscopy combined with N-way partial least squares discriminate analysis (NPLS-DA). Two brands of Mengniu (MN) and Sanyuan (SY) pure milk and adulterated milk with urea (0.2–20 mg.mL−1) were prepared. One-dimensional (1D) NIR spectra of all samples were collected at room temperature and 30 ℃−55 ℃ (5 ℃ interval). Synchronous 2D NIR correlation spectrum of each sample was calculated under the perturbation of temperature. For comparing, the discriminant models of MN brand, SY brand, and two-brands of adultetated milk were built based on 1D NIR spectra (room temperature), temperature-related three-dimensional (3D) NIR spectra, and sychronous 2D correlation spectra, respectively. For 1D NIR spectra, the discrimination accuracies of three models of MN, SY, and two-brands for unknown samples were 81.5%, 88.9%, and 85.2%, respectively. For temperature-related 3D spectra, the discrimination accuracies of three models for unknown samples were 96.3%, 96.3%, and 90.7%, respectively. For 2D correlation spectra, the discrimination accuracies of three models for unknown samples were 100%, 100%, and 98.1%, respectively. The results show that the proposed method can provide better discrimination results than 1D spectra and temperature-related 3D spectra.

White lupin (L. albus) has high potential interest as a high-protein food or feed crop. In addition, the oil of its seed has high quality for human nutrition. Crop improvement for these traits would profit of low-cost, NIRS-based evaluation methods that could be applied to large numbers of genotypes. The aim of this work was developing and assessing calibration models for NIRS prediction of these traits, envisaging analyses either on whole grain samples or on ground samples. Samples for the reference analyses were chosen by applying the Kennard-Stone algorithm to the whole set of spectra recorded from 2342 samples, both for seeds and flours. A group of 146 samples was selected to calculate calibration models based on chemical analyses for lipid and protein contents (Soxhlet extraction and Dumas method, respectively). After chemometric elaborations of the collected NIR spectra, with a repeated double cross-validation, the best results were obtained with lupin flours spectra for the estimation of protein content, using 4 LV and a mean centering as pretreatment, with performances which are good enough for breeding purposes (RPD = 3.30). Predictions were somewhat worse with lupin flours spectra for oil content, which attained RPD = 2.46 with 2 LV and first derivative and mean center as pretreatments. Results were less satisfying for predicting protein or oil content based on whole seed spectra, a non-destructive sample scenario of special interest for selection based on individual seeds.

Nitrogen and phosphorus are important nutrient measurement indicators for the slurry field application. Conventional wet chemical methods have been used as a routine detection way of total nitrogen (TN) and total phosphorus (TP). However, it is time-consuming, costly and destructive that cannot realize real-time and on-site detection. For the rapid and reliable determination of TN and TP in dairy farm slurry, near infrared spectroscopy (NIRS) was employed in this study. 472 samples were collected from 33 dairy farms in Tianjin. The near infrared diffuse reflectance spectra of all samples were scanned using Fourier transform near infrared spectrometer. And partial least squares models were established for quantitative analysis of TN and TP. Results were as follows: the correlation coefficient Rp were 0.92 and 0.91, the root mean square error of prediction (RMSEP) were 426.14 mg/L and 16.65 mg/L, the residual predictive deviation (RPD) were 2.73 and 2.63 for TN and TP, respectively. The prediction results of TN were better than that of TP. The results manifest that it is feasible to rapidly determine the contents of slurry TN and TP via the NIRS. This study can provide the technical support for reasonable land application of slurry.

The composition of milk powder (powdered milk) determines its quality and nutritional value. Currently, the standard or traditional methods that measure content of main components of milk powder have some disadvantages. In this study, ultraviolet (UV) spectroscopy combined with multivariate calibration/regression model was used to simultaneously predict the value of four main quality parameters including protein, fat, carbohydrate and energy rather than single component content in milk powder. Partial least squares (PLS) was chosen to establish regression model with the optimized number of principal factor. Without component separation/purification in the measurement with UV spectroscopy and pretreatment process in PLS modeling, good prediction results of multi-parameters were obtained with low root mean square error of prediction (RMSEP), high correlation coefficients (>0.98) and high RPD (Residual predictive deviation). By comparison, the results obtained by directly using work curve method were not satisfactory. Furthermore, PLS model acquired accurate and robust results than those of multivariate linear regression (MLR) model. It indicates that with the help of PLS, UV spectrometry is an effective, fast and simple “green” technique to simultaneously detect content of main parameters in milk powder. The proposed method could be applied to the quality control of milk powder, and be studied further to extend to quantitative analysis of milk liquid and even other food.

Soluble solids content (SSC) in apples was determined by a portable near-infrared (NIR) spectrometer combined with chemometrics methods. To build a stable partial least squares (PLS) model, spectral pretreatment and variable selection methods were considered in this work. The result showed that the best spectral pretreatment method was the combination of Savitzky-Golay smoothing, first-order derivative, autoscale, standard normal variate, mean center. Variable selection method competitive adaptive reweighted sampling (CARS) achieved the best performance. Our work could be a useful tool for the fruit grading and the post-harvest management.

The shelf life of ready-to-eat rocket salad packed under modified atmospheres was evaluated. Freshly cut rocket salad was packed in plastic bags under three modified atmospheres (A = atmospheric air; B = 30% O2, 70% N2; C = 10% CO2, 5% O2, 85% N2). At t = 0 and after 0, 1, 4, 7, 11 and 13 days NIR spectra were collected with a microNIR OnSite-W spectrometer (VIAVI Srl, Italy) in reflectance mode, over the range 900–1600 nm (50 scans; 125 reading points). Aquaphotomic approach was used to evaluate the maintenance of product freshness studying the changes in the water absorption profile. Samples were also analyzed by a Portable Electronic Nose PEN3 (AIRSENSE Analytics GmbH, Germany) with a sensor array composed of 10 metal oxide semiconductor (MOS) type chemical sensors. The PCA, applied 1300–1600 nm region, allowed the samples grouping according to the storage time. The obtained Aquagrams showed shifts of the selected water absorptions bands during the shelf life, estimating a first loss of freshness after 4–7 days from packaging for A and C theses, and after 7–11 days for B thesis. Similarly, E-nose detected important variations in MOS sensors data after 7 days for A and C theses, and after 11 days for the B one. NIR and E-nose results agreed in identifying the B modified atmosphere as the best for maintaining the product freshness. The B composition, characterized by a high O2 concentration, seemed to be able to lengthen by about 3 days the shelf life of the ready-to-eat rocket salad.

This study evaluated near infrared (NIR) spectra (n = 970) of nasal secretions (NS) from dairy calves (n = 5) challenged with bovine respiratory syncytial virus (BRSV). This pathogen is a common cause of respiratory disease in young calves worldwide and is typically diagnosed by evaluation of the clinical signs, followed by time-consuming serological and molecular methods. More rapid diagnostic methods could improve outcomes for infected calves. The near infrared aquaphotomics evaluation of this biofluid unveiled changes between the spectra (1300–1600 nm) of samples collected during the uninfected (n = 200) and infected (n = 200) stages, specifically identified in the WAMACS (water matrix coordinates) C1, C9, C10, and C11, where water molecules are highly associated with chaotropic solutes in water asymmetrical stretching vibrations (ν3) and with kosmotropic solutes in water clusters with 2, 3, and 4 hydrogen bonds (S2, S3, S4). These chemical differences were discriminated by PCA-LDA using a leave-one-animal-out approach with averaged percentages of accuracy, sensitivity, and specificity of 90.1 ± 4.3, 88.1 ± 3.8, 92.0 ± 5.5 in the calibration process, respectively. By collecting spectra from nasal secretions, we revealed the potential of NIR spectroscopy in combination with aquaphotomics and chemometrics for the detection of this viral infection in-vivo; as a first step toward developing a rapid in-field diagnostic tool for BRSV infection.

The research and development of fast, simple and accurate technology for breast cancer screening has important application value. In this paper, the discriminant analysis models for breast cancer and normal control samples were established using serum Vis-NIR spectroscopy combined with the equidistant combination-partial least squares-discriminant analysis (EC-PLS-DA) method. Standard normal variable (SNV) method was adopted for the spectral pretreatment of serum samples to improve spectral prediction.The parameters of the selected optimal EC-PLS-DA model were initial wavelength (I) = 1976 nm, ending wavelength (E) = 2396 nm, number of wavelengths (N) = 31, number of wavelength gaps (G) = 7 and the number of PLS latent variables (LV) = 10, respectively. In modelling, the calibration, prediction and total recognition accuracy rates were 96.0%, 97.5%, and 96.7%, respectively. Using independent validation samples not involved in modelling, the positive, negative, and total recognition accuracy rates were 85.0%, 90.0%, and 87.5%, respectively. The results showed the feasibility of serum Vis-NIR spectroscopic applied to discriminant analysis of breast cancer and normal control samples. The EC-PLS-DA method can extract information wavelengths, improve the recognition accuracy of discriminant analysis and reduce wavelength model complexity. The relevant wavelength model can provide valuable references for specialized spectrometer design and clinical application. The analytical technique is simple and novel, and has potential application in breast cancer screening.

Vis-NIR spectroscopy combined with equidistant combination-PLS (EC-PLS) method was applied for the rapid and reagent-free analysis of blood Hematocrit (HCT). The multi-parameter optimization platform based on Norris derivative filter (NDF) was constructed to select appropriate spectral preprocessing. Multi-partition modeling and independent validation in calibration-prediction-validation design were adopted to ensure the stability of parameter selection and the objectivity of modeling effect. For male and female groups, the optimal EC-PLS models of the grouping modeling were selected and achieved significantly better validation effects than hybrid modeling. In independent validation, the root mean square error of prediction (SEP) of male, female and mixed sample groups were decreased by 12.7%, 32.4% and 20.4%, respectively. The results showed that the predicted and clinical actual values of the all validation samples have high correlation coefficient of prediction (RP = 0.93) and low prediction error (SEP = 1.21%), and thus have potential for clinical application.

Recently, new low-cost multispectral sensors have been commercialized, paving the way for a large number of new agricultural applications (fertilization, grass cover, etc.), particularly for small farms. However, such sensors have never been tested for agricultural applications taking into account practical constraints (external environment, etc.). This study proposes to investigate the potentialities of the “AS7265x” chipset that presents a real interest for a wide range of applications in agriculture due to 18 spectral bands available. The first study involved the testing of three sensors in laboratory to assess the accuracy of the different spectral bands as well as the reproducibility of the measurements from one sensor to another but not presented here. In a second step, the work aimed at testing the potential of the sensor on real fields with two applications in proxy-detection to estimate the percentage of weeds over soil and the vine vigor. These field experiments focused on NDVI that is a vegetation index widely used in precision agriculture for proxy-sensing. Results show that although accurate, the sensors present some different bias for each wave bands and each sensor. These drawbacks require each sensor to be specifically calibrated before use which may limit their dissemination in agriculture. Once the sensor measurements are normalized, the NDVI values are consistent compared to the reference values given by the Greenseeker (R2 = 0.87 for NDVI < 0.75). Hence, the accuracy obtained was sufficient to differentiate the levels of grass cover and the differences in vegetative expression of the vine induced by local environmental effects.

The production process of Chinese medicine is characterized by complex processes, tedious steps and complicated influences, as all aspects of formulation production affect the final quality of a Chinese medicine product. Online NIR spectroscopy has the advantages of rapid and non-destructive, and can be used as an analytical technique for rapid evaluation of critical quality properties in the production process of traditional Chinese medicine. This paper systematically described the analysis and control methods of online NIR spectroscopy in the production process of traditional Chinese medicine from the application perspective of enterprises, and takes the online NIR spectroscopy analysis platform for traditional Chinese medicine built by Zeda Xingbang Pharmaceutical Technology Co., Ltd. as an example, to elucidate the feasibility of the application of online NIR spectroscopy in traditional Chinese medicine more comprehensively, discussed the economic benefits of the application of online NIR detection technology, and provided insights into the future The feasibility of the application of online NIR spectroscopy for Chinese medicine production is more comprehensively elucidated. The economic benefits of the application of on-line NIR detection technology were discussed and an outlook was also made for the future application of NIR technology in the field of Chinese medicine.

Process analysis can effectively stabilize pharmaceutical quality and optimize the control of production process. This study attempted to use a portable near-infrared spectroscopy for rapid detection of a Chinese medicine tablets from production line. First, PLS regression models were established for coating film at twelve different locations of the tablet section, and the results showed that the correlation coefficients of training and validation sets were all over 0.80. Subsequently, the twelve locations were divided into six groups to further establish regressions. After chemometrics optimization, the optimal of six group models were generally better than single location models, with Rc2 and Rv2 all above 0.85, and RMSEV values all below 2.0. The proposed approach can successfully realize on-site and online pharmaceutical monitoring and has a promising practical value.

To realize the rapid screening of industrial hemp for procurement, a method for evaluating the content of total cannabidiol (CBD) and total tetrahydrocannabinol (THC) of industrial hemp was established based on near infrared (NIR) reflectance spectroscopy. Both smashed un-decarboxylation industrial hemp samples and smashed decarboxylation industrial hemp samples were scanned. The spectral information was optimized by combining spectral pretreatment. These quantitative models were established based on partial least squares (PLS) and leave one cross validation. The content of total CBD and total THC models of industrial hemp were established on the condition of first derivative, standard normalization and de-trend pretreatment methods and the band range of 950–1650 nm. Through model comparison, it was found that sample pretreatment affects model accuracy. The calibration correlation coefficient (R2(c)), the cross validation root mean square error (RMSECV) and the prediction root mean square error (RMSEP) of the best total CBD model was 0.9803, 0.2888, 0.2425, respectively. The R2(c), the RMSECV and the RMSEP of the best total THC model was 0.9726, 0.0486, 0.0285, respectively. In the practice aplication, it could identify samples with high content CBD and avoid samples with high content THC (more than 0.3%). The handheld spectrometer could be used for the rapid determination of industrial hemp content.

The Near Infrared Spectroscopy (NIRS) was employed for control and monitoring of the tableting process during a continuous manufacturing process. The tableting production is key step in the production of solid dosage forms. Two spectrometers were embedded in the press and controlled by the press automation without any additional computer in order to have a fast and robust measurement. 72 batches were produced to calibrate and to validate the NIRS models at the BU and the CU position. A precise calibration (API content in %) have been obtained and a full speed acquisition was possible. The embedded spectroscopy for the control and the monitoring of the tableting process was demonstrated. The proposed strategy is an adequate Process Analytical Technology tool for continuous manufacturing and will enable opportunities for Real Time Release.

Polyester is an important textile material and the main part of waste textiles. Predicting the content of waste polyester fibers is the key to realizing the classification and recycling of waste textiles. In this paper, a total of 273 samples of polyester/nylon, polyester/wool, and polyester/cotton were used as the research targets, and quantitative analysis models of three types polyester blends were established by using near-infrared online analysis technology combined with partial least squares (PLS). The selection of preprocessing methods and the optimization of evaluation factors in the modeling process were also discussed. When the preprocessing method is Savitzky-Golay Derivative + Vector Normalization + Multiplicative Scatter Correction + Mean Centering, the quantitative analysis models of polyester/nylon and polyester/cotton blended fabrics predicted the best results. Among them, the number of evaluation factors for the polyester/nylon model is 9, and the evaluation factor for the polyester/cotton model is 5. When the Savitzky-Golay Derivative + Mean Centering was selected, and the number of evaluation factors is 6, the prediction effect of the polyester/wool model is the best. In this experiment, the built model was internally verified, and the accuracy of the model is higher than 95% under a tolerance of 3%. The model was tested externally using 90 polyester samples that were not involved in the modeling, and the overall prediction accuracy of the model was 94.4%. The established models can be applied to the quantitative prediction of three polyester blended fabrics.

Strong convolution of numerous overtones and combination bands makes NIR spectra difficult to interpret. Recent advances in anharmonic simulations decisively improved comprehension of NIR bands. Still, computational cost of accurate simulation remains very high, which hinders its wide use by non-specialized laboratories. In this proceedings article of NIR-2021 Conference, we discuss effective approaches to this problem, with optimizations for less time-consuming computations. Taking the example of piperine, the most popular spice ingredient in world trade, we directly compared two time-efficient approaches to this problem. The simulated NIR spectrum reveals an inherently complex structure with a large number of convoluted bands, mostly binary combinations, in particular in the 5500–4000 cm−1 range. The detailed assignments of NIR bands of piperine allowed to interpret the characteristics of the PLS regression models of the piperine in black pepper. Two models were compared, developed for spectral data sets obtained with the benchtop instrument (NIRFlex N-500) and a miniaturized spectrometer (microPHAZIR). These two spectrometers use different optical principles (benchtop: FT-NIR with a polarization interferometer and microPHAZIR: a programmable MEMS Hadamard mask), leading to profound instrumental differences. However, both are able to capture the most significant NIR absorption of piperine. In conclusion, the sensitivity of the two instruments to certain types of piperine NIR vibrations is different, with the stationary spectrometer being much more selective. This difference in capturing chemical information from the sample results in the difference in performance between the laboratory FT-NIR spectrometer and narrow-waveband miniaturized spectrometer in analyzing the piperine content in black pepper.

The multi-brand identification technology of wine has important application prospects. Since the main components of wine are roughly the same, and the characteristic components that can distinguish wine brands are usually trace amounts. The conventional quantitative detection method for brand identification is complicated and difficult. The naive Bayes classifier is an algorithm based on probability distribution, which is simple and particularly suitable for multi-class discriminant analysis. However, the absorbance probability between spectral wavelengths is not necessarily strongly independent, which limits the application of Bayes method in spectral pattern recognition. In this paper, a Bayes classifier algorithm based on wavelength optimization was proposed. First, a large-scale wavelength screening for equidistant combination (EC) was performed, and then wavelength step-by-step phase-out (WSP) was carried out to reduce the correlation between wavelengths and improve the accuracy of Bayes discrimination. The proposed EC-WSP-Bayes method was applied to the 5-category discriminant analysis of wine brand with visible and near-infrared (Vis-NIR) spectroscopy. The wavelength combination of the optimal EC-WSP-Bayes model was 412, 510, 1098, 1980, 2274, 2372 nm located in the visible, short-NIR and combination frequency regions. In modelling and independent validation, the total recognition accuracy rate (RARTotal) reached 97.6% and 98.7%, respectively. The technology is quick, easy, and has potential application in market. The proposed model of less-wavelength and high-efficiency (N = 6) can provide a valuable reference for small specialized spectrometer design. The proposed integrated EC-WSP-Bayes method can reduce the correlation between wavelengths, improve the recognition accuracy and applicability of Bayes method.

Based on model population analysis (MPA), the ensemble refinement (ER) has been proposed for outlier detection in calibration transfer. The ER first constructs many subsets of transfer set, and then computes the validation errors of each subset. After that, for each sample, the average error for subsets which include the one sample can be obtained. Finally, the samples with large average errors can be identified as outliers. The simulated dataset has been used to testing this method. The results showed that for calibration transfer methods such as CCA-ICE, DS and SST, ER can all identify outliers.

Firefly algorithm (FA) combined with partial least squares (PLS) are developed for near infrared (NIR) spectral interval selection and quantitative analysis of complex samples. The method firstly segments the near-infrared spectra into a number of intervals. Vectors with 1 and 0, which represent the interval selected or not, are used as the inputs of the FA. The RMSEP value predicted by PLS model is used as the fitness function of the FA. The number of spectral intervals, the population number, environmental absorbance and the constant of FA are optimized. With the optimal parameters, FA-PLS model is established and applied to predict protein, hemoglobin and cetane number in wheat, blood and diesel fuel samples, respectively. The results show that FA-PLS can significantly improve the prediction accuracy compared with full-spectrum PLS model.

The maturity of machine learning algorithms such as convolutional neural networks have made it possible to use them in feed spectroscopy. This paper compares convolutional neural network (CNN) and multivariate scattering processing support vector machine (MSC-SVM) modeling, including NIR spectroscopy, Raman spectroscopy modeling, and NIR-Raman spectroscopy modeling, to predict the protein content in feed. The experiments were based on measured NIR (wave number 4000–12000 wavenumbers) and Raman spectral (500–3000 wavenumbers) data due to the complementary roles of NIR and Raman spectroscopy techniques. The organic combination of the two spectral data adds useful information in model building. The CNN and MSC-SVM models based on NIR-Raman spectra, the predictions are better than single spectra.

Extreme learning machine (ELM) has received increasing attention in multivariate calibration of complex samples due to its advantages of fast learning speed and good generalization ability. However, irrelevant variables in spectral matrix to target can interfere the quality of ELM modeling. Therefore, variable selection is required before multivariate calibration. In this study, least absolute shrinkage and selection operator (LASSO) combined with ELM (LASSO-ELM) is used for spectral quantitative analysis of complex samples. In the method, LASSO is firstly used to selected variables by shrinking regression coefficients of unselected variables to zero. The optimal model position s of LASSO is determined by Sp criterion. Then ELM model is built between the selected variables and analyzed target with the optimal activation function and hidden node number determined by the ratio of mean to standard deviation of correlation coefficients (MSR). Near infrared (NIR) spectra of tobacco lamina and ultraviolet (UV) spectra of fuel oil samples are used to evaluate the prediction performance of LASSO-ELM. Results show that only with tens of variables, LASSO-ELM achieves the lowest root mean square error of prediction (RMSEP) and highest correlation coefficient (R) compared with full-spectrum partial least squares (PLS) and ELM. Thus, LASSO-ELM is an effictive variable selection and multivariate calibration method for quanatitive analysis of complex samples.

Grey Wolf Optimizer (GWO) algorithm, based on swarm intelligence, is easy to implement due to its few parameters and simple structure characteristics. However, few studies employ GWO for the spectral analysis to our knowledge. In this study, the GWO algorithm is introduced into the detection of nitrogen content to provide a reference for detecting nitrogen content in rubber leaves, which further facilitated the rubber yield of the rubber tree. GWO is used to optimize the SVR model of support vector regression, and 11 GWO-SVR models are established by taking spectral data of different fractional-order as input consecutively. The results show that the GWO-SVR model is superior to the SVR model as GWO-SVR optimizes SVR parameters by GWO algorithm, penalty factor c, and kernel function parameter g. The prediction correlation coefficient (Rp) improves by 10.88%, and the root mean square error (RMSEP) reduces by 9.15%. The GWO-SVR model base on fractional-order is optimized at 0.6 order, and the correlative Rp and RMSEP are 0.907 and 0.213, respectively. Compared with the GWO-SVR model based on the original spectrum, the Rp increases by 2.96%, and the RMSEP is reduced by 3.15%. Hence, it is feasible to predict rubber tree nitrogen content based on fractional GWO-SVR, which provides technical support for variable rate fertilization in precision rubber tree agriculture.

As the most important base for studying the quality stability of tobacco products, the characteristics of flue-cured tobacco are of great significance for both cigarette enterprises and producing areas. In this study, the characteristics of tobacco were qualitatively recognized based on the directly acquired gas chromatography-mass spectrometry (GC-MS) accumulative data via the pattern recognition, in an effort to rapidly and conveniently identify the grades information of tobacco. Specifically, an independent component analysis -back propagation neural network (ICA-BPNN) method was proposed to process the GC-MS ion accumulation. First, independent components were extracted after cumulating all the spectrum peaks of the acquired mass data. Next, a BPNN recognition model was established for the obtained independent components and then used to qualitatively discriminate four tobacco grades from Yunnan Province in China. Finally, a comparison was made between ICA-BPNN and principal component analysis (PCA)-BPNN models in the qualitative effect. Given nodes of 40, the ICA-BPNN model achieved the best accuracy, with the accuracy of calibration and prediction set of 72.86% and 80.82%, respectively. Results revealed that the proposed pattern recognition method, where mass data of tobacco were directly accumulated as overall information, is of certain potentials in the fast discrimination of agricultural product quality.

Dynamic light scattering (DLS), microscope, temperature-dependent infrared (IR) spectroscopy with perturbation correlation moving window (PCMW) were used to investigate the phase transition of a thermo-sensitive copolymer poly(N,N-dimethylacrylamide)-b-poly(hydroxypropyl acrylate)-b-poly(N,N-dimethylacrylamide) (PDMAA-b-PHPA-b-PDMAA). The copolymer was synthesized via reversible addition-fragmentation chain transfer (RAFT) polymerization. In the DLS analysis, the sudden increase in particle size at 49 ℃ and the polymer droplets appearing at 49 ℃ in the microscope photos indicate that the lower critical solution temperature (LCST) of the copolymer is around 49 ℃. IR spectra show that the OH groups of PHPA change from hydration states to free ones. The redshift of the spectra in C-H bands indicates the dehydration of CH groups. The hydration C = O bands transform into free states or the other part, which owns stronger intermolecular interactions. Moreover, the O-H bands show a more responsive to temperature during the phase transition by PCMW analysis. The reduced hydrophilicity of O-H bands is not enough to stabilize the polymer in water, which leads to a phase transition.