INTRODUCTIONNear-infrared spectroscopy (NIRS) is an increasinglypopular technique used for non-destructivequality evaluation in a variety of industries, includingthe food, agricultural, pharmaceutical, and woodindustries [1–3]. It ensures rapid and easy measurementswithout the need for multiple chemical reagents. RecentNIRS methods include online measurement, portablemeasurement, and imaging analysis [4–6]. NIRS iscontinuously expanding its uses in food analysis andbecoming an important tool for food quality control.The quality of coffee as a beverage is determinedby multiple factors such as the production system,geographical origin, chemical composition of roastedbeans, and final beverage characteristics. Raw coffeebeans contain a wide range of chemical compoundswhich interact amongst themselves at all stagesof coffee roasting, resulting in greatly diverse finalproducts [7–9]. For instance, the caffeine content, whichhas a significant effect on the final quality of coffeeproducts, needs to be determined fast and reliably byanalytical techniques.Wet chemistry methods are traditionally used toevaluate coffee quality and chemical characteristics,but these methods are destructive and time-consuming.Therefore, it is in scientific interests to find rapid, moreobjective, and simpler analytical methods for routinecoffee analysis to replace the old methods.Recent research has shown that spectroscopy in nearinfrared(NIR) and mid-infrared (MIR) radiation isuseful in coffee analysis [10–20]. Infrared spectroscopy(especially NIRS) coupled with chemometrics hasbeen proposed as an analytical method to determinethe degree of coffee roasting, adulterants in groundcoffee, and sensory attributes [17, 18, 21]. It is also usedto distinguish between robusta and arabica varieties,296Wójcicki K. Foods and Raw Materials. 2022;10(2):295–303discriminate coffee based on origin, and predict itschemical composition [15, 20, 22–24].The growing global demand for specialty coffeeincreases the need for improved coffee qualityassessment. For this reason, Tolessa et al. proposed NIRspectroscopy to predict specialty coffee quality [13].They examined the NIR spectra of 86 green Arabicabean samples of various quality. To create a modelthat correlates spectral data to cupping score data,they applied the partial least squares (PLS) regressionmethod. The high correlation coefficient between themeasured and predicted cupping scores (R2-values of90, 90,78, 72 and 72) indicate that NIR spectroscopycoupled with chemometric analysis could be a promisingtool for fast and accurate prediction of coffee quality andfor classifying green coffee beans into different specialtygrades.The sensory analysis of espresso coffee withthe attenuated total reflectance-Fourier transforminfrared spectroscopy (ATR-FTIR) was proposed byBelchior et al. [10]. The authors evaluated the potentialof ATR-FTIR and chemometrics in discriminatingespresso coffees with different sensory characteristicsreported by a panel of coffee tasters. They performedpartial least-squares discriminant analysis (PLS-DA)based on spectroscopic data to classify the coffeesamples according to their sensory qualities,demonstrating the potential of FTIR and chemometricanalysis in assessing coffee quality.In another study, Magalhaes et al. proposed FT-NIRspectroscopy and PLS regression as a non-destructiveand rapid tool to assess the content of three mainphenolics (caffeic acid, (+)-catechin, and chlorogenicacid) and methylxanthines (caffeine, theobromine, andtheophylline) in spent coffee grounds [11]. The best PLSmodel was obtained for caffeine content (0.95) followedby caffeic acid (0.92), (+)-catechin (0.88), theophylline(0.84), and chlorogenic acid (0.71), indicating FT-NIRspectroscopy as a suitable technique to screen spentcoffee grounds.Mees et al. identified coffee leaves using FT-NIRspectroscopy and soft independent modelling by classanalogy (SIMCA) [12]. In particular, they investigatednine taxa of Coffea leaves harvested over nine yearsin a tropical greenhouse of the Meise Botanic Garden(Belgium). The FT-NIR coupled with SIMCA allowedthe authors to discriminate the spectral profile bytaxon, aging stage, and harvest period with a correctclassification rate of 90, 100, and 90%, respectively.NIRS, PLS, and variable selection were used byRibeiro et al. to predict concentrations of a wide rangeof compounds in raw coffee beans [15]. The authorsproposed NIR spectroscopy coupled with chemometricsas a low-cost, rapid, and eco-friendly method in bothoff-line and on-line analyses of coffee beans and coffeebeverages. The obtained values of root mean squareerror of prediction (RMSEP) (0.08, 0.07 and 0.27) andrcv (0.98, 0.96, and 0.96) showed linear relations ofPLS models for quantifying caffeine, trigonelline, and5-caffeoylquinic acid, respectively.Near-infrared spectroscopy was used by Macedo et al.to evaluate the chemical properties of intact green coffeebeans based on PLS regression models [25]. The highestdetermination coefficients obtained for the samplesin the validation set were 0.810, 0.516, 0.694, and 0.781for moisture, soluble solids, total sugar, and reducingsugars, respectively. These results indicate that theNIR technology can be applied routinely to predict thechemical properties of green coffee.In another study, Baqueta et al. investigated theuse of NIR spectroscopy in conjunction with the PLSapproach to identify the sensory properties of coffee [21].The coffee samples varied in species, production region,variety, drying conditions, transit, postharvest procedure,storage times, coffee blend, coffee composition, androasting process. The performance of PLS models wasverified with the following merit parameters: sensitivity,accuracy, linearity, residual prediction deviation, fit,quantification, and detection limits. Since all thesensory qualities were predicted with acceptable valuescompatible with the merit criteria, the created modelswere suitable for quantifying, detecting, differentiating,and predicting the sensory features of coffee samples.Kyaw et al. reported encouraging findings aboututilizing NIR spectroscopy to forecast the moisturecontent of ground unroasted coffee beans [26]. Thespectral data processed with second derivative andKubelka-Munk (K/S) data yielded good accuracy formoisture prediction (r = 0.87 and accuracy = 99%).In view of the above, we aimed to develop a simple,rapid, and accurate method for evaluating the qualityof coffee samples by NIR spectroscopy, especially toinvestigate changes in the coffee spectra during roasting.STUDY OBJECTS AND METHODSSamples. Our study objects were arabica coffeesamples roasted by the Cafe Creator in Poznań, Poland.The coffee samples were divided into four groups basedon their origin, namely (1) Brazil, (2) Guatemala, (3)Peru, and (4) Congo. Their roasting parameters includedthe roaster power and roasting time (Table 1).Color measurements. The color of 41 samplesof coffee beans was measured by the L* a* b* methodusing a Konica Minolta Chroma Meter CR-310trichromatic colorimeter. Each sample was measured10 times. Before the measurements, the device wascalibrated against a white standard with the followingparameters: Y = 93.00, x = 0.3170, y = 0.3330. The entireanalysis was carried out using a D65 light source, i.e. thedaylight phase and the CIE L* a* b* color system.Near-infrared (NIR) measurements. NIR spectrawere performed on a MPA/FT-NIR spectrometer(Bruker). Single beam spectra of the coffee sampleswere collected and rationed against the background ofair. For each sample, the NIR spectra were recordedfrom 12500 to 400 cm–1 by co-adding 16 interferogramsat a resolution of 4 cm–1. Each sample was measured five297Wójcicki K. Foods and Raw Materials. 2022;10(2):295–303times. The coffee was ground in an electric grinder formeasurements. Between the measurements, the sampleswere mixed in order to obtain reliable results. Themeasurements were registered in the OPUS software(Bruker, USA).Partial Least Squares (PLS) regression. The PLSregression method was used to determine relationsbetween the spectra and the roasting time. Independentvariables (X) were the NIR spectra and dependentvariables (Y) were the color parameter or the roastingtime. Full cross-validation was applied to the regressionmodel. The regression models were evaluated using theadjusted R2 and the root mean-square error of crossvalidation.The quality models were evaluated by theratio of the standard deviation of reference data to theroot mean-square error of prediction, or the ratio ofperformance to deviation. The PLS analysis was carriedout using the Unscrambler X software (CAMO, Oslo,Norway).RESULTS AND DISCUSSIONColor measurements. Table 2 shows the colormeasurements of coffee beans in the L* a* b* system.The L* parameter is responsible for the brightnessof color in the tristimulus model. The higher it is, thegreater the brightness of the tested sample. Among thecoffees under study, the green coffee beans from Peruhad the highest L* value, i.e., the highest brightness.The Congo coffee, which was roasted at the power of80% for 12 min, had the lowest L* parameter, i.e., thelowest brightness. All the samples had positive a* andb* values, with their shades varying between red andyellow.As we can see in Table 2, the green coffee beansshowed the greatest brightness, followed by the samplesroasted for 8 min. With the increasing degree of roasting,the color of coffee beans became darker, which isconsistent with literature [27, 28].Spectral characteristics of coffee samples. Figure 1shows the absorption spectra of the coffees from Brazil,Congo, Guatemala, and Peru roasted for 12 min (80%roasting power). The spectral range was recordedthroughout the region of 12500–4000 cm–1. The mostintense absorption bands were recorded in the rangeof 8230–4440 cm–1. The spectra were characterized byseven bands with maximum absorption at 8238, 6819,5800, 5700, 5100, 4700, and 4440 cm–1. These bandscorresponded to the C-H, N-H, and O-H vibrations [29].The spectral range of 4545–4000 cm–1 corresponded tothe C-H stretching vibrations. The bands in the regionof 5000–4545 cm–1 were assigned to the combinationof the N-H and O-H stretching vibrations. The rangeof 6060–5555 cm–1 corresponded to the first tone of theC-H stretching vibration. In the 7142–6666 cm–1 region,it was associated with the first shade of the N-H andO-H stretching vibrations, while the absorption bandin the 7692–7142 cm–1 range was derived from the C-Hstretching vibrations. The band in the region of 9090–8163 cm–1 originated from the second tone of the C-Hstretching vibrations [30]. Specific chemical compoundscan be described with the following wavenumbers:caffeine (8865, 7704, 5981, 5794, and 5171 cm–1),trigonelline (8865 cm–1), chlorogenic acid (6770, 5794,5171, and 4699 cm–1), lipids (6770, 5794, 5171, and 4699cm–1), hydrocarbons (6770, 5171, and 4699 cm–1), sucrose(5794, 5405, and 5171 cm–1), proteins and amino acids(5171 cm–1), and water (5171 cm–1) [9, 14, 31]. Table 3presents the origin of the bonds occurring at the givenwavenumbers for the tested coffee beans.Coffee roasting. Many physical and chemicalchanges take place during coffee roasting. The methodof roasting depends on the origin of coffee beans andconsumer preferences. Heavily roasted coffee has alower nutritional value than light coffee [32].Numerous efforts have already been made touse NIR spectroscopy as an alternative technique todetermine coffee quality during roasting and analyze itschemical composition. According to Ribeiro et al., NIRspectroscopy can be used to determine the relationshipTable 1 Roasting parameters of coffee samplesOrigin Power of the roaster, % Roasting time, minBrazil Green –75 8, 10, 12, 1380 8, 10, 1295 8, 10, 12Guatemala Green –75 8, 10, 12, 1580 8, 10, 12, 1495 8, 10, 12, 13Peru Green –75 8, 10, 12, 1480 8, 10, 12, 1495 8, 10, 12, 14Congo Green –80 8, 10, 12Figure 1 Absorption spectra of ground coffee in near-infraredregion (12500–4000 cm–1)Absorption1.11.00.90.40.30.50.60.70.812000 11000 10000 9000 8000 7000 6000 5000 4000Wavenumber, cm–1GuatemalaPeruBrazilCongo298Wójcicki K. Foods and Raw Materials. 2022;10(2):295–303Table 2 Color measurements of green and roasted coffee beansOrigin Power of the roaster, % Roasting time, min L* average b* average a* averageGuatemala (green beans) – – 51.670 ± 0.20 0.870 ± 0.120 11.700 ± 0.141Guatemala (roasted beans) 95 8 45.440 ± 0.163 5.980 ± 0.057 12.650 ± 0.01310 42.410 ± 0.233 5.240 ± 0.064 9.810 ± 0.17712 38.050 ± 0.099 3.550 ± 0.099 6.250 ± 0.08513 36.390 ± 0.318 3.210 ± 0.042 4.760 ± 0.08680 8 45.480 ± 0.255 6.060 ± 0.156 12.290 ± 0.38310 42.460 ± 0.283 5.300 ± 0.106 9.980 ± 0.16312 39.720 ± 0.191 4.150 ± 0.077 7.380 ± 0.20514 36.770 ± 0.282 3.230 ± 0.071 4.550 ± 0.08575 8 48.120 ± 0.184 6.280 ± 0.163 14.470 ± 0.13410 43.210 ± 0.269 5.600 ± 0.106 10.490 ± 0.18512 40.620 ± 0.184 4.570 ± 0.099 7.860 ± 0.12015 36.460 ± 0.141 3.300 ± 0.064 4.160 ± 0.141Peru (green beans) – – 53.440 ± 0.042 0.880 ± 0.049 12.840 ± 0.078Peru (roasted beans) 95 8 42.540 ± 0.120 5.560 ± 0.085 10.430 ± 0.14810 38.940 ± 0.099 4.330 ± 0.064 7.280 ± 0.05712 38.060 ± 0.410 3.710 ± 0.121 6.440 ± 0.23414 36.860 ± 0.155 3.390 ± 0.064 5.270 ± 0.04980 8 45.010 ± 0.057 6.360 ± 0.064 12.620 ± 0.04210 42.240 ± 0.078 5.510 ± 0.092 10.420 ± 0.12712 39.910 ± 0.156 4.160 ± 0.020 7.510 ± 0.10614 37.350 ± 0.120 3.200 ± 0.049 5.360 ± 0.09975 8 45.350 ± 0.092 6.440 ± 0.057 12.930 ± 0.10610 40.240 ± 0.157 4.850 ± 0.085 8.290 ± 0.14212 39.270 ± 0.099 4.120 ± 0.049 7.130 ± 0.07114 37.610 ± 0.134 3.530 ± 0.021 5.440 ± 0.071Congo (green beans) – – 51.440 ± 0.134 0.610 ± 0.085 11.050 ± 0.071Congo (roasted beans) 80 8 42.150 ± 0.141 4.620 ± 0.041 9.270 ± 0.07810 39.240 ± 0.141 3.810 ± 0.078 6.680 ± 0.07812 36.120 ± 0.156 2.670 ± 0.085 3.950 ± 0.041Brazil (green beans) – – 52.360 ± 0.205 1.140 ± 0.057 13.080 ± 0.058Brazil (roasted beans)958 46.240 ± 0.092 6.810 ± 0.064 13.910 ± 0.09910 41.650 ± 0.128 5.360 ± 0.640 9.310 ± 0.06412 36.980 ± 0.085 3.460 ± 0.085 4.780 ± 0.064808 45.600 ± 0.071 6.040 ± 0.084 13.110 ± 0.09910 42.290 ± 0.134 5.800 ± 0.099 10.290 ± 0.06512 37.930 ± 0.092 3.910 ± 0.071 5.860 ± 0.042758 43.570 ± 0.184 6.190 ± 0.057 11.300 ± 0.06410 39.340 ± 0.128 4.600 ± 0.057 7.340 ± 0.01412 37.440 ± 0.170 3.710 ± 0.058 5.160 ± 0.08513 36.690 ± 0.134 3.080 ± 0.057 4.210 ± 0.071between the quality of a coffee cup and the chemicalcomposition of roasted coffee beans [9]. In addition, theauthors created a model from roasted beans to predictthe quality attributes of a coffee cup (e.g. acidity, body,and flavor).The relationship between some coffee roastingvariables (weight loss, density, and moisture) andnear-infrared spectra of original green and differentlyroasted coffee samples was investigated byAlessandrini et al. [14]. They developed separatecalibration and validation models based on partialleast square (PLS) regression, correlating NIR spectraldata of 168 representatives and suitable green androasted coffee samples with each roasting variable.As a result, the authors constructed robust and reliablemodels to predict roasting variables for unknownroasted coffee samples, considering that measured vs.predicted values showed high correlation coefficients(0.92–0.98).Pires et al. used multivariate calibration and NIRspectroscopy to correctly predict roasting degreesin ground coffee and coffee beans as a substitute forthe Agtron method [18]. The mathematical modelsfor predicting Agtron values of new coffee samplesusing the PLS approach were based on the associationbetween NIR spectra data and Agtron reference results.All Agtron roasting characteristics were investigatedin order to create representative models. With RMSEP299Wójcicki K. Foods and Raw Materials. 2022;10(2):295–303Table 3 The origin of bonds occurring at given wavenumbersfor tested coffee beans [31]Bond type Wavenumber,cm–1CH3; second overtone; stretching symmetric 8545–8042CH 7020–6562CH3; first overtone; stretching asymmetric 5841–5751CH2; first overtone; stretching asymmetric 5725–5654OH; stretching 5234–5000CH; stretching 4954–4509CH3; stretching 4358–4302values of 4.48 and 3.67, respectively, the proposedmodels showed promising results in predicting roastingcharacteristics in roasted whole coffee beans and groundcoffees.Yergenson and Aston investigated the use of in situNIR spectroscopy in the prediction of cracking events(start and end) during coffee roasting in order to developa more robust method of roasting based on cracks [33].Two sets of popping sounds (first and second cracks) thatoccur during coffee roasting are essential indicators forestablishing the roasting endpoint. The coffee sampleswere roasted using various time-temperature profiles.In situ NIR spectroscopy proved to be a reliable tool inforecasting the start and finish times of first and secondcrack occurrences based on the PLS regression (PLSR)with audio recordings from coffee roasting.The NIR spectra of coffees (beans and ground)roasted under different conditions are shown in Fig. 2.The obtained spectra were similar to each other,although varying in intensity. Longer roasting timelowered the intensity of the bands in all the ranges.This was due to decreased values of coffee components,as well as their volume and weight [34–36]. We foundthat the samples with the shortest roasting time(8 min) showed the highest absorbance, while those withthe longest roasting time (12 min) showed the lowestabsorbance at the same wavelength. We also noticedthat the intensity of the spectrum bands decreasedwith increasing roasting time. The NIR spectraobtained during the roasting assays were similar to thespectra reported in other studies [37, 38]. Accordingto the authors, the main changes in the spectra of theroasting process were an absorbance decrease in thewater band region (5200–5000 cm−1), which was dueto moisture loss, and an absorbance increase in thecombination band region (5000–4000 cm−1). A moredetailed discussion of the main wavelength intervalsand their relationships to chemical and physical changesin coffee during roasting can be found in the work bySantos et al. [37]. Our results were also consistent withthose reported by Catelani et al. [38]. The roastingprocess degraded coffee compounds, namelychlorogenic acid, coffee sugar, fat, and water. Literaturedata shows that the roasting time also affects thecaffeine content in coffee [39]. The longer the coffeeis roasted, the lower its caffeine content. All thesamples showed a lower intensity with an increase inthe roasting time. We concluded that regardless of theorigin, the roasting time caused a decrease in the coffeecomponents. The most intense bands occurred in thecoffees roasted for the shortest time, which means thatthey lost the least of their components and nutritionalvalue.The partial least squares (PLS) analysis wasperformed to determine the time of roasting. ThePLS models were obtained for the entire spectralrange (12500–4000 cm–1) and sub-regions withoutmathematical transformations (Table 4).We found good correlations between the spectraand the roasting time for all the coffee samples. TheR2 values for the calibration and validation curveswere 0.94 and 0.78, respectively. The root mean-squareerrors (RMSE) were low – 0.39 and 0.76 for calibrationand validation, respectively. The obtained models wereimproved when analyzing each type of coffee samplesseparately. Also, the sub-regions were used to improvethe model quality.There was a weak correlation between the spectraand the roasting power for all the coffee samples. Forthis reason, we analyzed the samples separately. Themost accurate model for Guatemala coffee was obtainedin the spectral region of 6813–5332 cm–1. The R2 was0.97 for calibration and 0.64 for validation. For Perucoffee, the spectral range of 5374–4954 cm–1 gave thebest quality model, with R2 values of 0.97 and 0.84 forcalibration and validation, respectively. There was nocorrelation between the spectra and the roasting powerfor Brazil coffee. The coffee from Congo was notanalyzed (only one power condition – 80%).The degree of coffee roasting can be assessed by thecolor: the longer the roasting, the darker the beans. Westudied a possibility of estimating the roasting time onthe basis of the NIR spectra by using the PLS analysisto correlate the NIR spectra (coffee beans) with theL* parameter (Table 4). By analyzing the values of thecalibration (R2 = 0.96) and validation (R2 = 0.95) curves,as well as the RMSE values (0.92 for calibration and1.05 for validation), we assumed that the coffee roastingtime could be determined based on the PLS regressionanalysis and the brightness parameter (L*).Our study indicates the potentiality of NIRspectroscopy in evaluating coffee quality. Based on thechanges of spectra, it is possible to monitor changesduring roasting. Chemometric analysis also deliveredvery promising results. The PLS models (for roastingtime and power conditions) hold potential as a rapidand reliable method which could be helpful in coffeemanufacturing. Our next step will be to determine thechemical composition of the coffee samples and identifythe potential of NIR spectroscopy in correlating roasting300Wójcicki K. Foods and Raw Materials. 2022;10(2):295–303conditions (time and power) with the chemical changesin order to select optimal roasting conditions for the finalproduct.CONCLUSIONOur study aimed to apply near-infrared spectroscopyto evaluate the quality of the coffee samples from Brazil,Guatemala, Peru, and Congo. We investigated theircomposition based on the spectral bands and vibrations.The regression results confirmed the possibilityof applying the NIR spectra to predict the roastingconditions. There was a correlation between the spectraand the roasting time, with the R2 of 0.94 and 0.78 forcalibration and validation, respectively. The RMSEswere low – 0.39 and 0.76 for calibration and validation,respectively. We also obtained a linear relation betweenthe spectra and the roasting power. The quality of themodels differed based on the coffee’s origin and subregion.All the coffee samples showed a good correlationbetween the spectra and the brightness (L* parameter).The R2 values were 0.96 and 0.95 for the calibration andvalidation curves, respectively.The results proved that NIR spectroscopy coupledwith chemometrics could be a promising tool to predictFigure 2 Changes in near-infrared spectra in coffee roasted at different power and time: (a) Guatemala coffee, (b) Peru coffee,(c) Brazil coffee, (d) Congo coffee. Full range spectrum (12500–4000 cm–1)Absorption1.11.00.30.60.712000 11000 10000 9000 8000 7000 6000 5000 4000b0.50.40.80.9Absorption1.21.00.40.60.812000 11000 10000 9000 8000 7000 6000 5000 4000aWavenumber, cm–1Guatemala 75% 10′Guatemala 75% 12′Guatemala 75% 15′Guatemala 80% 10′Guatemala 80% 12′Guatemala 80% 14′Guatemala 95% 10′Guatemala 95% 12′Guatemala 95% 13′Wavenumber, cm–1Peru 75% 10′Peru 75% 12′Peru 75% 14′Peru 80% 10′Peru 80% 12′Peru 80% 14′Peru 95% 10′Peru 95% 12′Peru 95% 14′Absorption1.11.00.40.712000 11000 10000 9000 8000 7000 6000 5000 4000cWavenumber, cm–1Brazil 75% 10′Brazil 75% 12′Brazil 75% 13′Brazil 80% 10′Brazil 80% 12′Brazil 95% 10′Brazil 95% 12′0.50.60.80.9dAbsorption1.11.00.40.712000 10000 8000 6000 4000Wavenumber, cm–1Congo 80% 10′Congo 80% 12′0.50.60.80.91.20.3 0.31.2301Wójcicki K. Foods and Raw Materials. 2022;10(2):295–303Table 4 Partial least squares (PLS) regression analysisPLS model Samples Spectral region, cm–1 Root mean-square error R2Calibration Validation Calibration ValidationRoasting time All coffee samples 12500–4000 0.39 0.76 0.94 0.78Guatemala 12500–40006813–53320.090.240.790.480.990.980.830.94Peru 12500–40006030–40000.390.180.700.820.940.990.850.80Brazil 12500–4000 0.22 0.30 0.96 0.95Congo – – – – –Roasting power All coffee samples – – – – –Guatemala 12500–40006314–52951.151.446.235.720.980.970.570.64Peru 12500–40006314–52956227–40005374–49544416–40901.701.260.841.512.296.164.324.543.844.270.960.980.990.970.930.580.790.770.840.80Brazil – – – – –Congo – – – – –Color (L* parameter) All coffee samples 12500–4000 0.92 1.05 0.96 0.95the roasting conditions of coffee samples. However, themodels developed in this study need to be further testedon independent data sets from other coffee varietiesto assess their stability and accuracy. Because of itscharacteristics, NIR spectroscopy has been applied indifferent production stages in the coffee industry: fromgreen coffee beans to the end product. The growinginterest in NIR spectroscopy is primarily due to thetechnique’s numerous advantages over other analyticaltechniques. In addition, this technique is nondestructiveand noninvasive, with a minimal or non-samplepreparation. NIR spectroscopy is also fast, low-cost,and robust, so it can be used in different environmentssuch as laboratories and industrial plants. In the future,the availability of portable instruments will also allowits use in the field. For these reasons, NIR spectroscopycould be named a “green technology”.CONFLICT OF INTERESTThe author declares that there is no conflict ofinterest.