1-s2.0-S0308814614010164-main

Food Chemistry 167 (2015) 264–271

Contents lists available at ScienceDirect

Food Chemistry

journal homepage: www.elsevier .com/locate / foodchem

Determination of sucrose content in sugar beet by portable visible andnear-infrared spectroscopy

http://dx.doi.org/10.1016/j.foodchem.2014.06.1170308-8146/Published by Elsevier Ltd.

⇑ Corresponding author. Address: 524 S. Shaw Lane, Room 224, Michigan StateUniversity, East Lansing, MI 48824, USA. Tel.: +1 517 432 8062.

E-mail address: [email protected] (R. Lu).

Leiqing Pan a, Qibing Zhu b, Renfu Lu c,⇑, J. Mitchell McGrath d

a College of Food Science and Technology, Nanjing Agricultural University, No. 1 Weigang Road, 210095 Nanjing, Chinab School of Communication and Control Engineering, Jiangnan University, 1800 Lihu Road, 214122 Wuxi, Chinac USDA Agricultural Research Service, Sugarbeet and Bean Research Unit, Michigan State University, 524 S. Shaw Lane, East Lansing, MI 48824, USAd USDA Agricultural Research Service, Sugarbeet and Bean Research Unit, Michigan State University, 1066 Bogue Street, East Lansing, MI 48824, USA

a r t i c l e i n f o a b s t r a c t

Article history:Received 26 February 2014Received in revised form 10 June 2014Accepted 29 June 2014Available online 5 July 2014

Keywords:Near-infrared spectroscopySugar beetSucrose contentDetectionHigh-performance liquid chromatography

Visible and near-infrared spectra in interactance mode were acquired for intact and sliced beet samples,using two portable spectrometers for the spectral regions of 400–1100 nm and 900–1600 nm, respec-tively. Sucrose prediction models for intact and sliced beets were developed and then validated. The spec-trometer for 400–1100 nm was able to predict the sucrose content with correlations of prediction (rp) of0.80 and 0.88 and standard errors of prediction (SEPs) of 0.89% and 0.70%, for intact beets and beet slices,respectively. The spectrometer for 900–1600 nm had rp values of 0.74 and 0.88 and SEPs of 1.02% and0.69% for intact beets and beet slices. These results showed the feasibility of using the portable spectrom-eter to predict the sucrose content of beet slices. Using simple correlation analysis, the study also iden-tified important wavelengths that had strong correlation with the sucrose content.

Published by Elsevier Ltd.

1. Introduction

Sugar beet is grown in a wide range of climatic conditions andin about 50 countries worldwide, including North America (UnitedStates and Canada), South America (Chile), Asia, North Africa (Mor-occo and Egypt), and most of Europe (Mosen, 2007). Sucrose con-tent is the most important trait in sugar beet production, and itis made up of more than 99.5% in the final white crystalline sugar.Hence breeders and researchers are striving to achieve high con-tent of sucrose in beet production, using genetic, molecular, andconventional breeding approaches.

Rapid measurement of sucrose content in sugar beets can assistbreeders in selecting promising germplasms and help sugar beetgrowers and processors in determining the yield and quality ofbeets after harvest and during storage and processing. Many meth-ods have been used to measure the sucrose content of beets,including polarimetry, enzyme-based spectroscopic assays, andhigh-performance liquid chromatography (HPLC) (McGrath &Fugate, 2012). Polarimetry is the generally accepted method usedin commercial sugar beet processing factories. Newer generationpolarimetric instruments can measure sucrose from dark andcoloured samples of molasses without juice clarification

(Singleton, Horn, Bucke, & Adlard, 2002). Enzyme-based spectro-scopic assays utilise a series of enzyme-catalysed reactions toquantitatively couple sucrose reaction to the synthesis of a spec-trally detectable compound (Spackman & Cobb, 2002). A relativelyrapid and inexpensive enzymatic-fluorometric microtitre plateassay was developed for sucrose quantification (Trebbi &McGrath, 2004). The method provided accurate and sensitivesucrose measurements from the tissues of young sugar beet rootswith a coefficient of determination (r2) of 0.976, but it was lessaccurate for older, field-grown root tissues (r2 = 0.605). Owing toits high sensitivity and specificity, HPLC is also used in the sucroseanalysis of sugar beet. But the technique is time-consuming andlabour intensive in sample preparation and sequential analysis(�12 min per sample) (Mulcock, Moore, Barnes, & Hickey, 1985).

Numerous studies have been reported in recent years on usingvisible and near-infrared (Vis/NIR) spectroscopy for the spectralregion of 400–2500 nm for fast measurement of soluble solids con-tent and other quality attributes of apple (Fan, Zha, Du, & Gao,2009; Mendoza, Lu, & Cen, 2012), peach (Carlomagno, Capozzo,Attolico, & Distante, 2004), pear (Xu, Qi, Sun, Fu, & Ying, 2012),pineapple (Chia, Abdul Rahim, & Abdul Rahim, 2012), Satsumamandarin (Gómez, He, & Pereira, 2006), sweet cherry (Lu, 2001),and many other fruits (Camps & Christen, 2009; Cayuela &Weiland, 2010; McGlone, Jordan, Seelye, & Martinsen, 2002; Paz,Sánchez, Pérez-Marín, Guerrero, & Garrido-Varo, 2009; Wang,Nakano, & Ohashi, 2011). NIR reflectance spectroscopy was used

L. Pan et al. / Food Chemistry 167 (2015) 264–271 265

to determine sucrose content in ground soybean samples (Sato,Zahlner, Berghofer, Lošák, & Vollmann, 2012). NIR spectroscopywas reported for achieving excellent predictions of sucrose in choc-olate, with a correlation coefficient of 0.997 or higher (Da CostaFilho, 2009). Vis/NIR spectroscopy was also reported for measuringthe sucrose content of sugarcane. Mat Nawi, Chen, Jensen, andMehdizadeh (2013) applied visible and shortwave near-infrared(Vis/SWNIR) spectroscopy to predict the sugar content of sugar-cane based on skin scanning. They reported an r2 value of 0.91and a root mean square error of prediction of 0.721 �Brix. Roggo,Duponchel, and Huvenne (2004) measured the sucrose content ofmore than 2700 homogenised beet brei samples from 15 differentfactories using Vis/NIR (400–2500 nm) and reported excellentresults with a standard error of prediction of 0.10%.

While previous Vis/NIR studies have showed promising resultsfor sucrose measurement in beets, they needed to destroy andhomogenise beet samples. In addition, these studies were carriedout using expensive laboratory Vis/NIR instruments that coverthe entire visible and near-infrared region (i.e., 400–2500 nm).No research has been reported on measuring the sucrose contentof intact and sliced beets using spectroscopic instruments that onlycover the Vis/SWNIR region of 400–1100 nm or a portion of the NIRregion (i.e., 900–1600 nm). Spectral measurements for these twospectral regions offer considerable advantages over the full Vis/NIR region because they can be done more quickly and conve-niently, using low-cost portable or handheld CCD-based (charge-coupled device) (400–1100 nm) or InGaAs-based (indium galliumarsenide) (900–1600 nm) spectrometers.

Therefore, the objective of this research was to study the feasi-bility of Vis/SWNIR and NIR spectroscopy for predicting thesucrose content of intact and sliced sugar beets. The specific objec-tives were to: (1) measure spectra of intact and sliced sugar beetsusing two portable spectrometers operated in interactance mode,for the spectral regions of 400–1100 nm and 900–1600 nm, respec-tively; (2) develop statistical models for the spectral data to predictthe sucrose content of intact beets and beet slices; and (3) identifyimportant wavelengths that have strong contributions to thesucrose content prediction.

2. Materials and methods

2.1. Samples

Sugar beets used for this experiment were harvested from theMichigan Sugar Company’s official variety trials in the experimen-tal field of Michigan State University’s Saginaw Valley Researchand Extension Center at Frankenmuth, Michigan during the 2012harvest season. The beet samples came from different commercialhybrid varieties, and they were stored in refrigerated storage at4 �C for about 2 months prior to the experiment. Only those beetsthat did not show physiological deterioration (e.g., rot) and physi-cal damage (i.e., cuts and bruises) were selected for the study. Thesamples were moved from cold storage within 24 h prior to theexperiment and were washed to remove adhered soil. Spectralmeasurements were first made on intact beets using two portablespectrometers for the spectral regions of 400–1100 nm and 900–1600 nm, respectively (see details in the following section). There-after, the beet samples were cut into two sections in the directionthat is approximately perpendicular to the crown-root axis. Spec-tral measurements were immediately made at the centre of thelower section of the beet. After the spectral measurements hadbeen completed, wet lab chemistry analysis (i.e., HPLC) was per-formed to measure the sucrose content of each beet sample. A totalof 398 beet samples were tested in the study.

2.2. Visible and near-infrared spectroscopic measurement

Two portable spectrometers were used for this research: a Vis/SWNIR spectrometer (Model LOE-USB; tec5USA Inc., Plainview,NY) for the spectral region of 400–1100 nm, and an NIR spectrom-eter (Model NIR 512L-1.7T1; Control Development Inc., SouthBend, IN) for the spectral region of 900–1600 nm. Both spectrome-ters were operated in interactance mode (Fig. 1a). The sugar beetsample was illuminated with a broadband light source that wasdelivered from the 25-mm diameter ring of a lighting/detectionprobe, which was in contact with the beet sample. In the centreof the probe was a light detector covering an area of 11 mm diam-eter to receive the light that had travelled through the flesh tissue.A separating distance of 3.5 mm between the light illuminatingring and the detector ensured that only the light that travelledthrough the beet tissue would be received by the probe (Fig. 1b).A soft rubber sealing ring was used between the illuminating ringand the detector to block the illuminating light from entering thedetecting area directly. In addition, a sponge ring (5 mm thick)was attached to the periphery of the probe to block ambient light.

For Vis/SWNIR measurements, the lamp power supply was setto 100 W and the integration time of the spectrometer was set to575 ms. For NIR measurements, the lamp power supply was setto 200 W and the integration time was set to 4 s. Spectral measure-ments were made at a location approximately 10 mm from thecrown end of each intact beet. For both spectrometers’ measure-ments for intact beets, three scans were acquired from the samelocation of each beet sample, and they were then averaged for fur-ther analysis.

After spectral measurements for the intact sample had beenmade, the beet was sliced in the transverse direction (i.e., perpen-dicular to the crown-root axis) into two sections using a stainlesssteel kitchen knife. Spectral measurements were then made, usingeach spectrometer, from the lower section (toward the root side) ofthe beet. Two scans were taken at two locations that were approx-imately equidistant from the centre and the outside edge of eachbeet slice, and these values were then averaged. The integrationtimes were 0.2 s and 1.5 s for Vis/SWNIR and NIR measurements,respectively. Thereafter, one cylindrical specimen of 50 mm indiameter and 10 mm in height (without skin) was taken fromone of the two detection locations for the beet slice and subse-quently used for sugar content analysis by HPLC. Reference spectrawere acquired from a white Teflon disk in order to calculate therelative interactance of each beet sample.

2.3. Sucrose measurement by HPLC

Sucrose content was measured by means of high-performanceliquid chromatography (HPLC) (Trebbi & McGrath, 2004). Each beetspecimen of 30 g fresh weight was lyophilised until the pressurewas <1 mTorr for at least 3 h and then ground to fine powder withmortar and pestle. Pulverised dried tissue (100 mg) was resus-pended in 4 mL of 80% ethanol in a 5-mL fluted-cap tube (USA Sci-entific, Inc., Ocala, FL) and placed horizontally on an orbital shaker(50 rpm) at 40 �C for 16 h. The suspension was centrifuged at3000g for 10 min to obtain the clarified ethanol sugar extract. Analiquot (1.0 mL) of clarified ethanol sugar extract was vacuumdried, the pellet was resuspended in 1.0 mL of high-resistivitywater (18 MX cm–1), and the solution was passed through a0.22-lm nylon filter (Spin-X Centrifuge Tube Filter; Corning, NewYork, NY). The aliquot of water-resuspended sugar extract(1.0 mL) was used for HPLC analysis with a 6.5 mm � 300 mmWaters Sugar-Pak I carbohydrate column (WAT085188; WatersCo., Milford, MA). The mobile phase was 134 lM Na2CaEDTA at aconstant flow of 0.5 mL min–1, 90 �C, 12 min run time, and quanti-fied with a Waters 410 differential refractometer held at 35 �C,

Fig. 1. Schematic of the visible and near-infrared spectral measurement system (a), and the configuration of the lighting/detection probe (b).

266 L. Pan et al. / Food Chemistry 167 (2015) 264–271

according to the manufacturer’s instructions (Dorschel, 1984).Concentration standards for sucrose (2.92–46.74 mM) were usedto generate standard curves. Fig. 2 shows a typical HPLC chromato-gram, in which sucrose is clearly separated from other compoundspresent in the beet. The linearity for sucrose was established byplotting the peak area (y) versus concentration (x), which wasexpressed as: y = 661321x + 335005 (r2 = 0.9985). System controland data management were accomplished using Empower Chro-matography Manager software (Waters Co., Milford, MA).

2.4. Spectral treatment and model development

Relative interactance was calculated from the original sample,dark and reference spectral data, using the following equation:

RI ¼ ðIS � IDÞ=ðIR � IDÞ ð1Þ

where RI = relative interactance; IS = measured light intensity fromsample; IR = measured light intensity from the white Teflon refer-ence; and ID = dark current from the spectrometer. Like in reflec-tance or transmittance measurement, absorbance for interactancemay be calculated as log(1/RI). However, our preliminary analysisshowed that there was no advantage of using absorbance for devel-oping calibration models. Hence, this paper only presents the proce-dure of, and results from, analysing relative interactance.

Two standard data preprocessing methods, i.e., standard normalvariate (SNV) (Herrero Latorre, Peña Crecente, García Martín, &Barciela García, 2013; Huang, Zhao, Chen, & Zhang, 2014) and thefirst order Savitzky–Golay derivative (SG1) (Sirisomboon, Tanaka,Fujita, & Kojima, 2007), were applied to reduce multi-collinearityand the baseline offset arising from scattering effects, thus enhanc-ing the information related to chemical constituents. A partial leastsquares (PLS) model was developed by using MATLAB 7.5.0 (The

Fig. 2. HPLC chromatogram for analysis of sucrose in sugar

MathWorks, Inc., Natick, MA) with the PLS Toolbox (EigenvectorResearch, Inc., Wenatchee, WA).

All 398 sugar beet samples were first sorted for sucrose contentin ascending order. The samples were then systematically sepa-rated into two sets with 75% for calibration and the remaining25% for prediction (i.e., every fourth sample was taken out for pre-diction). A calibration model for sucrose prediction was developed,using partial least squares, for the calibration samples only. Theoptimum number of latent variables was determined by selectingthe first minimum from the predicted residual sum of squares(or PRESS) curve, calculated using the leave-one-out cross valida-tion method. After the calibration model had been developed, itwas then used to predict the sucrose content of beet samples inthe prediction set that were not used in the model calibration. Sta-tistical parameters, including correlation coefficient of calibration(rc) and prediction (rp), the standard error for the calibration(SEC) and prediction (SEP) data sets and the ratio of sample stan-dard deviation to standard error of prediction (RPD), were calcu-lated. These parameters were used to assess the performance ofeach calibration model for predicting the sucrose content of beets.

Since calibration and prediction results are affected, to someextent, by the sampling procedure, the above model calibrationand prediction procedure was repeated four times to ensure thatthe reported results more accurately reflect the actual performanceof the two spectrometers. As thus, after completion of the first runof calibration and prediction, another 25% samples were taken outfrom the calibration set and put aside as a new set of predictionsamples in the second run, while the prediction samples that wereused in the first run were returned to the calibration set. The sameprocedure of model calibration and prediction as that used in thefirst run was then followed in the second run. This procedurewas repeated four times until each sample in the dataset was

beet with the peak at 8.010 min representing sucrose.

L. Pan et al. / Food Chemistry 167 (2015) 264–271 267

eventually used once for prediction (Cen, Lu, Mendoza, & Ariana,2012). Finally, values of the statistics (i.e., rc, rp, SEC, SEP andRPD) from the four runs were averaged and reported.

3. Results and discussion

3.1. Spectral features for sugar beets

Fig. 3 shows the relative interactance spectra of beet slices for400–1100 nm and 900–1600 nm, respectively. Spectra for intactbeets were similar in pattern to those for beet slices.

Absorption in the visible spectral region (400–700 nm) wasweaker than that in the NIR region, since relative interactancegenerally decreased with wavelength over the spectral region of700–1000 nm (Fig. 2a and b) and over the spectral region of1100–1400 nm (Fig. 2b). This could be due to the fact that the colourof skin and flesh for sugar beets was close to white, meaning thatless light was absorbed by the tissue in the visible region. Vis/NIRspectra are sensitive to the organic compounds that are composedof the molecular bonds of C–H, O–H, and N–H. Sucrose is composedof C–H, O–H, C–C, and C–O. Two relatively small absorption peaks

Fig. 3. Relative interactance spectra of beet slices obtained using the visible andshortwave near-infrared spectrometer for 400–1100 nm (a) and the near-infraredspectrometer for 900–1600 nm (b).

at 480 nm and 637 nm, which are associated with chlorophyll b,were observed from the interactance spectra (Fig. 3a). For the spec-tral region of 700 nm–1100 nm, absorption mainly came from thethird overtone and second overtone of both C–H and O–H. Anabsorption peak around 980 nm (Fig. 3a) was likely due to the exis-tence of the O–H stretching vibration and water (Vanoli et al.,2011). Similar spectral profiles between 900 and 1100 nm wereobserved for both Vis/SWNIR and NIR instruments. Lower energyfor the spectral region of 1150–1400 nm could be due to the exis-tence of the second overtone and combination of first overtones ofC–H. Little energy was detected for the region of 1400–1600 nm,resulting from strong absorption of light by the beet tissue, dueto the combination of the first overtones of bond C–H and O–Hin H2O.

3.2. Sucrose content prediction

Sucrose content values for the 398 beet samples obtained fromthe HPLC analyses were approximately normally distributedaround the mean value of 20.43% (standard deviation = 1.48%)and ranged between 15.08% and 25.36%.

Different spectra pretreatment methods affected the predictionperformance of PLS models for sliced samples, as measured by rc

and SEC, rp and SEP, and RPD (Table 1). For the Vis/SWNIR spectra,SNV showed the best prediction results among the four spectralpretreatments (i.e., no pretreatment, SNV, SG1, and SNV + SG1),For the NIR spectral data, the three pretreatment methods pro-duced better results compared to those for no pretreatment.SNV + SG1 gave better predictions over the other pretreatments(Table 1). Both Vis/SWNIR and NIR spectra, after SNV andSNV + SG1 pretreatments respectively, provided acceptable predic-tions for the sucrose content of beet slices, even though NIR gener-ally performed slightly better (Table 1 and Fig. 3). For beet slices,the best correlation for the calibration model was 0.93 and theSEC was 0.55 for Vis/SWNIR, compared with the best correlationof 0.92 and the SEC of 0.59 for NIR. In predicting sucrose contentfor the prediction set of beet slices, the correlations (i.e., rp) were0.88 and 0.88, the SEPs were 0.70 and 0.69, and RPDs were 2.13and 2.14, for Vis/SWNIR (Fig. 4a) and NIR (Fig. 4b), respectively.

For intact beets, the three spectra pretreatment methods forboth Vis/SWNIR and NIR spectra produced mixed results forsucrose prediction, compared to that for no pretreatment (Table 2).For the Vis/SWNIR spectra, SNV + SG1 achieved better results thanno pretreatment, SG1 and SNV, whereas for the NIR spectra, SNVwas the best among the four pretreatments. The results in Tables1 and 2 demonstrate that although spectra pretreatment generallyimproved the PLS model prediction, no single pretreatmentmethod performed consistently better than the others. Further-more, the effectiveness of a pretreatment method was influencedby type of instrument (i.e., Vis/SWNIR versus NIR) and sample con-dition (intact versus sliced).

The results for intact beets are not as good as those for beetslices (Table 1 versus Table 2). With the selected spectra pretreat-ment method of SNV for NIR and of SNV + SG1 for Vis/SWNIR,the values of rc and SEC were 0.87 and 0.73 for Vis/SWNIR, whichare better than that for NIR (rc = 0.85 and SEC = 0.78). When themodel was used to predict sucrose content for intact beets, rp,SEP and RPD deteriorated for both Vis/SWNIR and NIR; the correla-tions were 0.80 and 0.74, the SEPs were 0.89 and 1.02, and the RPDswere 1.69 and 1.47 for Vis/SWNIR and NIR, respectively. Consistentwith the calibration results, Vis/SWNIR performed better than NIRin predicting sucrose content for intact beets. One possible expla-nation is that beet peel had more influence on the NIR measure-ment of intact beets than Vis/SWNIR measurement, as a result ofstronger absorption by the beet tissue in the NIR region than inthe Vis/SWNIR region. Alternatively, since the concentration of

Table 1Calibration and prediction results of sucrose content for beet slices using visible and shortwave near-infrared (Vis/SWNIR) and near-infrared (NIR) instruments with differentspectra pretreatments.*

Instrument Treatment rc SEC (%) rp SEP (%) RPD

Vis/SWNIR No 0.93 0.55 0.87 0.76 1.98SNV 0.93 0.55 0.88 0.70 2.13SG1 0.92 0.57 0.85 0.80 1.89SNV + SG1 0.92 0.58 0.86 0.76 1.95

NIR No 0.91 0.60 0.79 0.96 1.77SNV 0.93 0.55 0.86 0.76 1.98SG1 0.91 0.60 0.86 0.76 2.00SNV + SG1 0.92 0.59 0.88 0.69 2.14

‘No’ stands for no pretreatment for spectra; ‘SNV’, standard normal variate; ‘SG1’, the first order Savitzky–Golay derivative.The statistical values presented in the table are the average of four calibration and prediction runs.

* rc and rp are the correlation coefficients of calibration and prediction, respectively; SEC and SEP, standard errors for calibration and prediction, respectively; RPD, the ratioof sample standard deviation to standard error of calibration and prediction.

Table 2Calibration and prediction results of sucrose content for intact beets using the visible and shortwave near-infrared (Vis/SWNIR) and near-infrared (NIR) instruments with differentspectra pretreatments.*

Instrument Treatment rc SEC (%) rp SEP (%) RPD

Vis/SWNIR No 0.88 0.70 0.74 1.04 1.45SNV 0.87 0.72 0.64 1.37 1.25SG1 0.87 0.73 0.78 0.96 1.56SNV + SG1 0.87 0.73 0.80 0.89 1.69

NIR No 0.86 0.75 0.72 1.05 1.42SNV 0.85 0.78 0.74 1.02 1.47SG1 0.82 0.84 0.71 1.09 1.37SNV + SG1 0.85 0.78 0.73 1.03 1.45

‘No’ stands for no pretreatment for spectra; ‘SNV’, standard normal variate; ‘SG1’, the first order Savitzky–Golay derivative.The statistical values presented in the table are the average of four calibration and prediction runs.

* rc and rp are the correlation coefficients of calibration and prediction, respectively; SEC and SEP, standard errors for calibration and prediction, respectively; RPD, the ratioof sample standard deviation to standard error of calibration and prediction.

268 L. Pan et al. / Food Chemistry 167 (2015) 264–271

sucrose in the beet is not uniform and generally is higher in thecentre of the beet, deeper illumination via Vis/SWNIR might havereturned more representative spectra. It was difficult to achievesatisfactory sucrose content prediction for intact beets using eitherVis/SWNIR or NIR measurement in interactance mode. In additionto the presence of beet peel, the surface of intact beets wasunsmooth and irregular, which could, in turn, have affectedconsistent measurements of interactance spectra, thus yieldingpoorer Vis/SWNIR and NIR prediction of sucrose content. In somecases such as to assess sucrose content of in-ground sugar beetsprior to harvesting, it would be more desirable to measure sucrosecontent in intact beets. To meet such application needs, furtherimprovement to the lighting/detection configuration is needed.

It should be pointed out that during postharvest storage, phys-iological activities continue in beets, which would change some ofthe chemical constituents as well as the cellulosic structure. Thesephysiological changes would, in turn, affect Vis/NIR spectral mea-surements. Hence the calibration models developed in this studyfor beets that had been kept in cold storage for 2 months maynot be suitable for predicting freshly harvested beets or beets thathave undergone different postharvest storage treatments. Todevelop a more robust sucrose prediction model, further researchis needed for quantifying the effect of storage condition (i.e., tem-perature and time) on the sucrose prediction model.

3.3. Spectra correlation analysis

In the above results, contributions of individual wavelengths tothe prediction accuracy were not considered. Further analysis onthe correlation between the sucrose content and interactance forindividual wavelengths would enable us to ascertain the most use-

ful wavelengths for determining the sucrose content and thus gaina better understanding of light interaction with the beet tissue.Fig. 5 shows the correlation between the sucrose content and eachsingle wavelength in the Vis/SWNIR region of 400–1100 nm usingthe optimum spectra pretreatment for beet slices and intact beets,as determined from simple correlation analysis. For beet slices, thesucrose content was negatively correlated with interactance over400–650 nm and was positively correlated with interactance over650–1100 nm (Fig. 5a). The correlations were relatively constantand high, ranging between 0.3 and 0.5 for the spectral region of650–1100 nm, and between �0.5 and �0.3 for the spectral regionof 400–650 nm. No particular wavelengths showed a distinctlyhigher correlation (Fig. 5a). This is not totally surprising, in viewof reported studies for other plant products. For instance, Lu(2001) reported that the correlation between soluble solids con-tent and individual wavelength for sweet cherry was relativelysmooth and stable within the spectral region of 800–1300 nm,without showing a distinctly higher correlation at any specificwavelengths. However, the pattern of correlation over the NIRregion was quite different from that for the Vis/SWNIR region. Inthe NIR region of 900–1600 nm, several single wavelengths werestrongly correlated with sucrose content (Fig. 5c). The highest cor-relation of 0.6 occurred at 1080 nm, followed by the negative at980 nm. For the region between 1380 nm and 1600 nm, correla-tions were relatively constant and low, around –0.15, probablydue to the exceptionally low signal in this region resulting fromstrong light absorption by the peel (as shown in Fig. 3b). It shouldbe pointed out that different correlation patterns over the samespectral region of 900–1100 nm were noticed for the two spectrom-eters, which indicates the varying effects of spectra pretreatmentmethods and optical instruments on the spectral measurementand subsequent prediction of sucrose content for beets.

Fig. 4. Sucrose content predictions for beet slices using the visible and shortwave near-infrared (400–1100 nm) (a) and near-infrared (900–1600 nm) (b) spectrometers.

L. Pan et al. / Food Chemistry 167 (2015) 264–271 269

For intact beets, simple correlation spectra were quite differentfrom those of beet slices. There were several wavelengths thatshowed a strong correlation with sucrose content. The correlationswere higher at the wavelengths of 460 nm (r = –0.28), 660 nm(r = 0.36), and 1000 nm (r = –0.18) for Vis/SWNIR (Fig. 5b). ForNIR, there were two high correlation peaks of 0.2 at 930 nm and–0.33 around 1200 nm, and several lower peaks at other wave-lengths. The fact that some wavelengths had relatively strong cor-relations with the sucrose content suggested the possibility ofusing a few selected wavelengths to predict the sucrose contentof intact beets. This should be investigated in further research.

Correlation analysis demonstrated that type of instrument anddifferent spectral region directly affected the pattern of correlationbetween sucrose content and wavelength. Furthermore, the pat-tern of correlation was also dependent on whether spectral mea-surements were made on sliced or intact beets. These findingsare not totally unexpected, because plant materials are a complexbiological entity composed of multiple chemical constituents, withheterogeneous structural characteristics, all of which would havevarious degrees of influence on spectral measurements and, hence,

the correlation pattern. These results also show the challenge andneed of developing a general, reliable calibration model for accu-rate prediction of the sucrose content of beets that have undergonedifferent postharvest storage conditions.

4. Conclusion

The potential of two portable spectrometers for 400–1100 nmand 900–1600 nm was evaluated for sucrose content assessmentof intact and sliced sugar beets. Results showed good correlationbetween sucrose content and interactance spectra for beet slices,with a correlation (rp) of 0.88, standard error of prediction (SEP)of 0.70, and a ratio of sample standard deviation to standard errorof prediction (RPD) of 2.13 for Vis/SWNIR, and rp = 0.88, SEP = 0.69,and RPD of 2.14 for NIR. PLS models for Vis/SWNIR gave better pre-dictions of sucrose content for intact beets with rp = 0.80,SEP = 0.91 and RPD = 1.69, compared with rc = 0.72, SEP = 1.02 andRPD = 1.47 for NIR. Hence Vis/SWNIR and NIR in interactance modecan be used to predict sucrose content for beet slices with acceptable

Fig. 5. Simple correlation between the sucrose content and relative interactance for individual wavelengths of the visible and shortwave near-infrared for beet slices (a) andintact beets (b), near infrared for beet slices (c) and intact beets (d) after spectra pretreatment.

270 L. Pan et al. / Food Chemistry 167 (2015) 264–271

accuracy. However, further improvement in the sensing configura-tion and calibration methods is needed, in order to achieve satis-factory sucrose measurement for intact beets. Several importantwavelengths were identified, which showed relatively strong cor-relations with sucrose content, for both Vis/SWNIR and NIR mea-surement. Different spectral preprocessing and modellingmethods and postharvest storage effect need to be considered inorder to improve the sucrose content prediction accuracy for beetslices and intact beets.

Acknowledgements

The authors would like to thank Thomas Goodwill and R. ScottShaw in the USDA Agricultural Research Service (ARS) Sugarbeet &Bean Research Unit at East Lansing, Michigan, for their technicalsupport for the sucrose measurement by HPLC. The experimentwas carried out when the first and second authors were visitingscientists to the USDA/ARS research unit at Michigan State Univer-sity, East Lansing, Michigan. The research was also supported bythe Chinese National Foundation of Natural Science (31101282)and Special Fund for Agro-scientific Research in the Public Interest(201303088).

References

Camps, C., & Christen, D. (2009). Non-destructive assessment of apricot fruit qualityby portable visible-near infrared spectroscopy. LWT – Food Science andTechnology, 42(6), 1125–1131.

Carlomagno, G., Capozzo, L., Attolico, G., & Distante, A. (2004). Non-destructivegrading of peaches by near-infrared spectrometry. Infrared Physics & Technology,46(1), 23–29.

Cayuela, J. A., & Weiland, C. (2010). Intact orange quality prediction with twoportable NIR spectrometers. Postharvest Biology and Technology, 58(2), 113–120.

Cen, H., Lu, R., Mendoza, F. A., & Ariana, D. P. (2012). Assessing multiple qualityattributes of peaches using optical absorption and scattering properties.Transactions of the ASABE, 55(2), 647–657.

Chia, K. S., Abdul Rahim, H., & Abdul Rahim, R. (2012). Prediction of soluble solidscontent of pineapple via non-invasive low cost visible and shortwave nearinfrared spectroscopy and artificial neural network. Biosystems Engineering,113(2), 158–165.

Da Costa Filho, P. A. (2009). Rapid determination of sucrose in chocolate mass usingnear infrared spectroscopy. Analytica Chimica Acta, 631(2), 206–211.

Dorschel, C. (1984). Analysis of sugars 1: Retention times on Sugar-PakI. In WatersLab Highlight; LAH 0210, 2. Milford, MA: Waters Co..

Fan, G., Zha, J., Du, R., & Gao, L. (2009). Determination of soluble solids and firmnessof apples by Vis/NIR transmittance. Journal of Food Engineering, 93(4), 416–420.

Gómez, A. H., He, Y., & Pereira, A. G. (2006). Non-destructive measurement ofacidity, soluble solids and firmness of Satsuma mandarin using Vis/NIR-spectroscopy techniques. Journal of Food Engineering, 77(2), 313–319.

Herrero Latorre, C., Peña Crecente, R. M., García Martín, S., & Barciela García, J.(2013). A fast chemometric procedure based on NIR data for authentication ofhoney with protected geographical indication. Food Chemistry, 141(4),3559–3565.

Huang, L., Zhao, J., Chen, Q., & Zhang, Y. (2014). Nondestructive measurement oftotal volatile basic nitrogen (TVB-N) in pork meat by integrating near infraredspectroscopy, computer vision and electronic nose techniques. Food Chemistry,145, 228–236.

Lu, R. (2001). Predicting firmness and sugar content of sweet cherries using near-infrared diffuse reflectance spectroscopy. Transactions of the ASAE, 44(5),1265–1274.

Mat Nawi, N., Chen, G., Jensen, T., & Mehdizadeh, S. A. (2013). Prediction andclassification of sugar content of sugarcane based on skin scanning using visibleand shortwave near infrared. Biosystems Engineering, 115(2), 154–161.

McGlone, V. A., Jordan, R. B., Seelye, R., & Martinsen, P. J. (2002). Comparing densityand NIR methods for measurement of Kiwifruit dry matter and soluble solidscontent. Postharvest Biology and Technology, 26(2), 191–198.

L. Pan et al. / Food Chemistry 167 (2015) 264–271 271

McGrath, J. M., & Fugate, K. K. (2012). Chapter 30: Analysis of sucrose from sugarbeet. In V. R. Preedy (Ed.), Dietary sugars: Chemistry, analysis, function and effects(pp. 526–545). Cambridge, UK: Royal Society of Chemistry Publishing.

Mendoza, F. A., Lu, R., & Cen, H. (2012). Comparison and fusion of fournondestructive sensors for predicting apple fruit firmness and soluble solidscontent. Postharvest Biology and Technology, 73, 89–98.

Mosen, A. (2007). Beet-sugar handbook. New Jersey: John Wiley & Sons.Mulcock, A. P., Moore, S., Barnes, F., & Hickey, B. (1985). The analysis of sugars in

beet, part III: HPLC analysis. International Sugar Journal, 87, 203–207.Paz, P., Sánchez, M., Pérez-Marín, D., Guerrero, J. E., & Garrido-Varo, A. (2009).

Instantaneous quantitative and qualitative assessment of pear quality usingnear infrared spectroscopy. Computers and Electronics in Agriculture, 69(1),24–32.

Roggo, Y., Duponchel, L., & Huvenne, J. (2004). Quality evaluation of sugar beet (Betavulgaris) by near-infrared spectroscopy. Journal of Agricultural and FoodChemistry, 52(5), 1055–1061.

Sato, T., Zahlner, V., Berghofer, E., Lošák, T., & Vollmann, J. (2012). Near-infraredreflectance calibrations for determining sucrose content in soybean breedingusing artificial reference samples. Plant Breeding, 131(4), 531–534.

Singleton, V., Horn, J., Bucke, C., & Adlard, M. (2002). A new polarimetric method forthe analysis of dextran and sucrose. Journal of the American Society of SugarcaneTechnologists, 22, 112–119.

Sirisomboon, P., Tanaka, M., Fujita, S., & Kojima, T. (2007). Evaluation of pectinconstituents of Japanese pear by near infrared spectroscopy. Journal of FoodEngineering, 78(2), 701–707.

Spackman, V. M. T., & Cobb, A. H. (2002). An enzyme-based method for the rapiddetermination of sucrose, glucose and fructose in sugar beet roots and theeffects of impact damage and postharvest storage in clamps. Journal of theScience of Food and Agriculture, 82(1), 80–86.

Trebbi, D., & McGrath, J. M. (2004). Fluorometric sucrose evaluation for sugar beet.Journal of Agricultural and Food Chemistry, 52(23), 6862–6867.

Vanoli, M., Rizzolo, A., Grassi, M., Farina, A., Pifferi, A., Spinelli, L., et al. (2011). Time-resolved reflectance spectroscopy nondestructively reveals structural changesin ‘Pink Lady�’ apples during storage. Procedia Food Science, 1, 81–89.

Wang, J., Nakano, K., & Ohashi, S. (2011). Nondestructive evaluation of jujubequality by visible and near-infrared spectroscopy. LWT – Food Science andTechnology, 44(4), 1119–1125.

Xu, H., Qi, B., Sun, T., Fu, X., & Ying, Y. (2012). Variable selection in visible and near-infrared spectra: Application to on-line determination of sugar content in pears.Journal of Food Engineering, 109(1), 142–147.