NIR Spectroscopy Applications for Internal and External Quality Analysis of Citrus Fruit—A Review

REVIEW PAPER

NIR Spectroscopy Applications for Internal and ExternalQuality Analysis of Citrus Fruit—A Review

Lembe S. Magwaza & Umezuruike Linus Opara &

Hélène Nieuwoudt & Paul J. R. Cronje & Wouter Saeys &

Bart Nicolaï

Received: 25 February 2011 /Accepted: 26 September 2011# Springer Science+Business Media, LLC 2011

Abstract The global citrus industry is continually con-fronted by new technological challenges to meet the ever-increasing consumer awareness and demand for quality-assured fruit. To face these challenges, recent trend inagribusiness is declining reliance on subjective assessmentof quality and increasing adoption of objective, quantitativeand non-destructive techniques of quality assessment. Non-destructive instrument-based methods are preferred todestructive techniques because they allow the measurementand analysis of individual fruit, reduce waste and permitrepeated measures on the same item over time. A wide

range of objective instruments for sensing and measuringthe quality attributes of fresh produce have been reported.Among non-destructive quality assessment techniques,near-infrared (NIR) spectroscopy (NIRS) is arguably themost advanced with regard to instrumentation, applications,accessories and chemometric software packages. This paperreviews research progress on NIRS applications in internaland external quality measurement of citrus fruit, includingthe selection of NIR characteristics for spectra capture,analysis and interpretation. A brief overview on thefundamental theory, history, chemometrics of NIRS includ-ing spectral pre-processing methods, model calibration,validation and robustness is included. Finally, futureprospects for NIRS-based imaging systems such as multi-spectral and hyperspectral imaging as well as opticalcoherence tomography as potential non-destructive techni-ques for citrus quality assessment are explored.

Keywords Non-destructive evaluation . Near-infraredspectroscopy . NIRS . Citrus fruit . Internal quality . Externalquality . Hyperspectral . Multispectral . Optical coherencetomography (OCT) . X-ray computed tomography (CT)

Introduction

Citrus fruit are the highest value fruit crop in internationaltrade. Current annual worldwide citrus production isestimated at over 105 million tons, worth over US $20billion (Spreen 2009). Assessment of quality parameters isone of the oldest techniques in fruit postharvest manage-ment (Butz et al. 2005). Most acceptable citrus fruit qualityclassification systems, whether manually operated or

L. S. Magwaza :U. L. Opara (*)Postharvest Technology Research Laboratory,Department of Horticultural Science, Stellenbosch University,Stellenbosch 7602, South Africae-mail: [email protected]

U. L. OparaPostharvest Technology Research Laboratory,Department of Food Science, Stellenbosch University,Stellenbosch 7602, South Africa

H. NieuwoudtDepartment of Viticulture and Oenology,Institute for Wine Biotechnology, Stellenbosch University,Stellenbosch 7602, South Africa

P. J. R. CronjeCitrus Research International, Department of HorticulturalScience, Stellenbosch University,Stellenbosch 7602, South Africa

W. Saeys :B. NicolaïVCBT-MeBioS, Biosystems Department,Katholieke Universiteit Leuven,3001 Heverlee, Belgium

Food Bioprocess TechnolDOI 10.1007/s11947-011-0697-1

automated, are based on subjective assessment of visibleexternal aspects of fruit based on surface colour, size, shapeand presence of defects (Kader 2002; Sun et al. 2009). Thisis mainly because until recently only these attributes couldbe measured non-destructively at a speed compatible withthe typical speed of commercial sorting lines, which may beas high as 10 fruits/s (Nicolaï et al. 2008). Increasingconsumer demand for internal quality attributes such assweetness and nutritional content, coupled with industrydemand for innovative tools for rapid and cost-effectivedetection and monitoring of physiological disorders, havespurred considerable interest among researchers on theapplication of near-infrared spectroscopy (NIRS) on citrusfruit quality monitoring and evaluation. The objective ofthis paper is to discuss current knowledge about NIRspectroscopy applied to citrus quality analysis and identifyother emerging and innovative technologies for non-destructive investigation of external and internal qualityapplicable to citrus fruit.

Application of NIRS in Citrus Fruit Quality Analysis

Most of the NIRS research on citrus fruit have focused onassessing internal quality attributes (Liu et al. 2010a, b;Gómez et al. 2006; Fraser et al. 2003). Very limitedresearch work has been conducted on NIRS applicationsto assess, monitor and predict internal and rind physiolog-ical disorders (Cronje 2005, 2009). This section of thereview focuses on general aspects of NIRS applied to citrusfruit, including the selection of NIR characteristics andwavelength, data acquisition, analysis and interpretation.

Selection of NIR Characteristics

Typical visible–near-infrared (vis–NIR) spectra of citrusfruit, ‘Satsuma’ mandarin (Golic and Walsh 2006) and‘Valencia’ oranges (Magwaza et al. 2011) are shown in

Figs. 1 and 2, respectively. The pattern of the absorptioncurves is similar to that for other fruit such as kiwifruit(McGlone and Kawano 1998) and mango (Guthrie andWalsh 1997) although position and magnitude of the peaksare fruit specific. From the visible region (400–700 nm), acontinuous decrease in absorbance with the minimum at680 nm is observed. High absorbance observed at 672 nmis indicative of red absorbing pigments, particularlychlorophyll that gives the fruit its characteristically greencolour (Gómez et al. 2006). After this peak, there is a verysharp drop in absorbance as the spectrum enters the NIRregion. Gómez et al. (2006) estimated this drop to be 12-fold. From 680 to ∼910 nm, the absorbance spectrum staysrelatively flat until a prominent peak centred at 975 nmappears. This peak is most probably due to water andcarbohydrate since they absorb strongly at 958 (Williamsand Norris 1987; McGlone and Kawano 1998) and 935 nm(Kawano et al. 1993). Other two peaks related to the strongwater absorbance bands exist between 1,440 and 1,485 nm,and in citrus fruit these water absorption peaks occur at1,184 and 1,457 nm.

When working in the NIR region of the spectrum,knowledge of the infrared (IR) absorption wavelengths (3–1,000 μm) of certain functional groups is critical in knowingwhere to look in the NIR for overtones (Wetzel 1983).Compared with NIR, IR radiation is strongly absorbed byorganic molecules, with the wavelength of absorptioncharacteristic of the molecule. The absorbance in the NIRregion of the spectrum is typically 10–100 times weaker thanthat of the fundamental absorption bands in the IR region ofthe electromagnetic spectrum (Walsh et al. 2000). Weakabsorptions in the NIR have been cited as a shortcoming.

In contrast to IR peaks which are narrow and diagnostic,peaks in NIR spectra are broad, up to 100–150 nm wide

Fig. 1 A typical absorbance (log(1/R) spectrum of Satsuma mandarinin the wavelength from 400 to 2,350 nm (Gómez et al. 2006)

800 1000 1200 1400 1600 1800 2000 2200 2400

Wavelength (nm)

0.5

1.0

1.5

2.0

Abs

orba

nce/

Log

(1/

R)

977.5

1202.1

1450.3

1942.4

Fig. 2 Typical absorbance spectra of ‘Valencia’ oranges obtainedusing three different spectral acquisition modes, emission head(Matrix F) (blue solid line), fibre optic probe for solid samples(purple broken line) and integrating sphere (red broken line)(Magwaza et al. 2011)

Food Bioprocess Technol

(Walsh et al. 2000). Thus, NIR spectra are comprised ofbroad band which arise from overlapping absorptionscorresponding mainly to overtones and combinations ofvibrational modes involving C–H, O–H, N–H and S–Hchemical bonds (Osborne 2000). The molecular vibrations,particularly by O–H and C–H bonds, are responsible forstrong absorption bands in the NIR spectral region bybiological material, with the absorption band of water O–Hbonds dominating in hydrated material (Golic et al. 2003;Nicolaï et al. 2009).

Water constitutes about 80–90% of fruit and vegetables(Gómez et al. 2006) and because of the high water content,water absorption features (bands) dominate the NIRspectrum of fruit and vegetables (Palmer and Williams1974). The water absorption bands are fairly broad withcentres at approximately 970, 1,200, 1,450, 1,950 and2,250 nm (Williams and Norris 2001). The band assign-ments of the major water and sugar (O–H and C–H)vibrations are summarised in Table 1. Hydrated objects arecharacterized by complex hydrogen bonding interactionsbetween water, sugar, protein etc. In NIR spectra ofhydrated samples with large molecules such as fruit andvegetables, the effective absorption bands are relativelywide and complex, even at the fundamental (IR) frequency,

due to different chemical environment of each O–H and C–H bonds in water and sugar molecules (Guthrie et al.2005a). Wavelength drift or shift in informative peaks maybe due to differences in chemical composition, changes inthe optical path in the spectrometer and temperaturechanges (Gaffney 1973; Zude et al. 2008).

Data Acquisition Modes

A host of investigators have extensively evaluated thepotential to measure quality attributes of intact citrus fruitby means of NIRS (e.g. McGlone et al. 2003; Cayuela2008). However, application and data acquisition techni-ques vary among different groups of investigators. Inpractice, there are five different NIRS measurement modesfitting different applications. These modes are transmit-tance, interactance, transflectance, diffuse transmittance anddiffuse reflectance (Huang et al. 2008). Three types of dataacquisition configurations frequently used for intact citrusfruit quality evaluation and prediction in the NIR systemare reflectance, transmittance and interactance (Schaare andFraser 2000; Fraser et al. 2003; Cayuela and Weiland 2010).

Reflectance spectra provide information about theappearance attributes, which are commonly accepted

Table 1 The band assignmentsof the major vibration bonds(Golic et al. 2003)

Vibrational frequency overtones

Tentative assignment Fundamental 1st 2nd 3rd

OH stretching

nm 2,860–3,120 1,410–1,440 970 738

cm−1 3,200–3,500 6,950–7,100 10,300 13,550

OH combinations

nm 1,920–2,068 1,100 840

cm−1 14,800–5,200 9,090 11,900

CH stretching

nm 3,300–3,470 1,600–1,800 1,100–1,230 910

cm−1 2,880–3,000 5,550–6,250 8,100–9,100 11,000

CH combination

nm 2,100–2,352

cm−1 4,250–4,750

CH2 stretching

nm 3,460–3,500 1,720–1,765 1,215 930

cm−1 2,880–2,910 5,670–5,820 8,230 10,750

CH2 combination

nm 2,310–2,325

cm−1 4,300–4,330

OH, CH and CH2 deformations

nm 6,900–8,330 2,250–2,320 2,400–2,600 1,850–2,12011 111–25,000

cm−1 1,200–1,450 4,310–4,440 3,840–4,170 4,720–5,400


criteria for the product quality evaluation. Generally, thereflectance spectra do not contain a lot of information aboutthe internal quality of the flesh because of the limitedpenetration depth of NIR radiation into the fruit (Krivoshiev etal. 2000). Also, reflectance spectra may be confounded withspecular reflection. One of the advantages of reflectancemode as reported by Fu et al. (2007) is that measurementsare easier to obtain and the light levels of the reflectedradiation are much higher than that of transmitted radiation,however, variations in superficial and surface properties ofthe fruit may influence calibrations.

In interactance mode, the field of view of the detector isseparated from the illuminated surface by a light seal in contactwith the fruit surface (Schaare and Fraser 2000). Basically, therelative merit of interactance mode is that it provides acompromise between reflection and transmission modes.

Transmission mode is considered the most frequentlyused and appropriate to assess internal quality attributessuch as sugars and acids levels of thick-skinned fruit suchas mandarin and orange (Kawano et al. 1993; Miyamoto etal. 1998; Cayuela 2008). However, implementation oftransmittance spectra measurement mode is limited by twofundamental factors. The first is the fact that transmittancespectra of fruit with different sizes, thickness of peel andshape have a different optical path and optical density (OD)(Krivoshiev et al. 2000). It is therefore expected thatcalibration model performance of spectra collected usingtransmission optical geometry will be affected by size,shape and peel thickness (Guthrie et al. 2005a). Second,because of the low penetration depth of NIR radiation, theintensity of the incident NIR light needs to be very high tohave at least some transmitted. Such high light intensitylevels may cause permanent thermal damage to the fruit atthe illuminated spot.

Since the fruit peel is part of the light path, the spectrumof flesh optical density will vary depending on the changesin peel OD. Basically, fruit should be considered as anobject with three-layer structure of ‘peel–flesh–peel’ andthe distortion influence of the peel should be adjustedaccordingly during the analysis of transmittance spectra(Kawano et al. 1993). However, according to Dull et al.(1989a), it is accepted that for products of the same species andvariety the peel thickness is constant. In a study carried out bythe authors, it was concluded that the inclusion of the peel inthe light path does not change the fundamental dependenciesbetween the spectra and chemical data. A similar argumentwas reported in a research paper by Krivoshiev et al. (2000).This team of investigators argued that the transmittance of theflesh or internal quality can be predicted with sufficientaccuracy regardless of the peel spectrum.

In a later study of cantaloupe, Dull et al. (1989b) establisheda higher R value of 0.97 and lower standard error ofcalibration (SEC) of 0.56°Brix when determining SSC of

slices compared to when intact fruit were measured. On intactfruit measurements r decreased to 0.60 and SEC increased to1.67°Brix. In melons, Dull and Birth (1989) observed lower rvalue (0.87) and higher SEC value of 1.6°Brix. From thesestudies, it would seem that in fruit with thick peel, thedistorting influence of the peel is more pronounced than thosewith thin peel. This argument is in accord with the statementby Kawano et al. (1993) who noted that it would be moredifficult to determine internal composition in fruit with thickpeel than in fruit with thin peel. Interactance or bodytransmittance NIR spectrometric techniques are recommen-ded modes for the internal quality assessment of fruits andvegetables (Peiris et al. 1999).

McGlone et al. (2003) explored the relative merits ofthree different measurement modes of the NIRS on‘Sastuma’ mandarin fruit. The transmittance mode deliv-ered the most accurate SSC prediction (R2=0.96, RSMEP=0.32°Brix). Results on the interactance mode were superior(R2=0.85, RSMEP=0.47°Brix) to those of the reflectancegeometry (R2=0.75, RSMEP=0.63°Brix). Irrespective ofslightly higher prediction outcomes using transmittance thatwere reported by the authors, good results using reflectance(diffuse reflectance) mode have also been reported onoranges (Lu et al. 2006; Cayuela 2008; Cayuela andWeiland 2010; Liu et al. 2010a) and mandarin fruit (Gómezet al. 2006). Ou et al. (1997) and Greensill and Walsh(2002) also developed calibration models of mandarinsusing interactance geometry and reported very goodcalibration statistics. Another group of investigators thathas used interactance mode of NIR to assess internal qualityattributes of citrus fruit was Guthrie et al. (2005a).

Similarly, on kiwifruit, Schaare and Fraser (2000)evaluated the ability of three modes of vis–NIR spectro-scopic measurements to non-destructively estimate internalquality and reported the interactance mode to provide themost accurate estimates of SSC, density and flesh colour.SSC, for instance was predicted with a standard error ofprediction (SEP) of ±0.80°Brix and r value of 0.96. Fu etal. (2007) later compared transmission and reflectancemodes of vis–NIR spectroscopy for detecting brown heartdisorder in pears. The results obtained by this group ofinvestigators indicated that transmission mode was betterthan reflectance mode for internal disorder detection.

Wavelength Selection and Light Penetration Depth

NIR at different wavelengths ranges extending from thevisible spectrum to the NIR has been used and establishedby many researchers to define different quality parametersof fruit and vegetables. However, due to the inherentvariability in fruit types and the influence of growingenvironment, there is no consent between researchers on thebest wavelength range to study each fruit quality parameter


(Peiris et al. 1999). The wavelength region over whichKawano et al. (1993) made the NIR measurements forsugar content of ‘Satsuma’ mandarin was from 680 to1,235 nm and reported spectral absorption at 844 nm to berelated to fruit size. Similarly, Miyamoto and Kitano (1995)considered spectral absorption at 840–855 to be related tofruit diameter and found the key wavelengths for TSScalibration model on peeled and intact mandarin fruit to be770 and 905 nm, respectively. Miller and Zude (2002) usedNIRS in the region of 400–1,100 nm to develop models (r=0.82) to measure SSC level on grapefruit and tangerinefruit. McGlone et al. (2003) concluded that a wavelengthrange from 750 to 1,100 nm was optimal for internalquality assessment of mandarin fruit. In a study aimed atassessing internal quality attributes such as SSC and DM ofmandarin fruit, Guthrie et al. (2005a, b) used the wave-length range between 720 and 950 nm. In a research workto evaluate the use of vis–NIRS in measuring the firmness,SSC and acidity of Satsuma mandarin fruit, Gómez et al.(2006) developed sound models using the full vis–NIRspectral range (350–2,500 nm). However, spectral measure-ments below 500 nm are usually too noisy to use in the modelcalibration (McGlone et al. 2002). Lu et al. (2006) developedcalibration models to predict SCC (r=0.995, root meansquare error of prediction (RMSEP)=0.79°Brix) of ‘Gannan’citrus fruit in the full spectrum range of 800–2,500 nm. Liuet al. (2010a) developed a SSC prediction model in thewavelength range of 350–1,800 nm. In a later study, Liu etal. (2010b) selected the wavelength range between 350 and1,040 nm with a 0.2-nm sampling interval to develop SSCand TA prediction models of ‘Nanfeng’ mandarin fruit.

From the foregoing literature evidence, it is clear thatmajority of investigators used the short wave spectral rangeof the NIR radiation (700–1,100 nm) to measure citrus fruitinternal quality. A first reason is that this range iscompatible with that of cheap silicon detectors. Further,this window is relevant to sugar and water as it include thesecond and third overtone of OH stretching and vibrations.Correspondingly, there is a decreased risk of maskingspectral information correlated to low concentration ofconstituents. In addition, this range is advantageous for itslight can penetrate much farther in fruit of many species(Pallav et al. 2009). Guthrie and Walsh (1997) attempted toassess the depth of tissue that contributes to the NIRspectrum (reflectance mode) of intact fruit. They reportedthat the depth of penetration was dependent on thewavelength and concluded that all spectral informationfrom the fruit was derived from within 5 mm of the fruitsurface, with an exponential decrease in information withincreasing depth within the 5 mm zone. The penetrationdepth in apple fruit in the 700–1,100 nm window has beenreported to be from a few mm up to 25 mm (Lammertyn etal. 2000; Fraser et al. 2001,) depending on the data

acquisition mode. In fact, this wavelength range was shownto be readily transmitted through citrus fruit irrespective ofits thick rind (Fraser et al. 2003). Using illuminated laserlight with 808 nm wavelength Fraser et al. (2003) observeda rapid reduction in light levels across the thick skin(∼4 mm into the fruit) of mandarin fruit. Compared to theskin, they observed a less rapid reduction as the lightcontinued to pass into the flesh (∼14 mm into the fruit). Inthe tissue regions (∼22–32 mm into the fruit) they reportedan exponential decay and a rapid drop across the distal skin(∼50 mm into the fruit). The steep decline in light levels asit passes through the skin of the fruit shows that the citrusthick skin is highly attenuating compared to the flesh, ashas been asserted by Kawano et al. (1993).

Researchers have closely scrutinised the hypothesis thatprocessing the full spectrum can always lead to the bestresults. However, researchers such as Centner et al. (1996)and Carlini et al. (1999, 2000) argued that best suitedmodels often involve few tens of carefully selected wave-lengths. Centner et al. (1996) developed an eliminationprocedure, called uninformative variable elimination-PLSand tried it on chemical data sets. This method out-performed standard full spectrum PLS. Forward selectiontechnique for the identification of wavelengths to beincluded in a PLS regression (PLS) model has also beentried by Osborne et al. (1997, 1999) and the calibrationsobtained improved. Carlini et al. (1999) presented resultson the use of wavelength selection methods for the accuratedefinition of models for evaluation of SSC in fruit. Aselection scheme involving the sub-sampling of the spectralinterval from 600 to 1,100 nm, with a fixed step, wassystemically able to produce better final prediction modelsthan plain PLS. The interval-PLS (iPLS), which wasdeveloped by Nørgaard et al. (2000), searches for a spectralinterval that is particularly informative with respect to theparameter under consideration. The iPLS is based on theprinciple of splitting the spectra into smaller equidistantregions and the regression are developed in every subinter-val (Peirera et al. 2008). This method has been shown to beeffective on citrus, mandarin fruit in particular, by improv-ing PLS models significantly (Liu et al. 2010b). On anattempt to predict SSC and TA on ‘Nafeng’ mandarin fruit,this group of investigators assessed different wavelengthranges and the optimal (681.36–962.75 nm) was chosen onthe basis of the lowest root mean square error of cross-validation (RMSECV) and RMSEP values. This wouldseem logical as this NIR range is strictly known to containimportant carbohydrate, sugar and water absorbances.

Data Acquisition Positions

The distribution and level of attributes, such as TSS, withinthe fruit may differ with maturation, growing conditions


and size of the fruit (Guthrie et al. 2005a, b). According toPeiris et al. (1999), the coefficient of variation of TSSwithin a single orange fruit was 10.2%, 1.8% and 5.6% inproximal to distal ends, around the fruit circumference(equatorial) and radial orientation, respectively. A similartrend was reported by the authors on grapefruit. It is knownthat TA differs with fruit positions in citrus (Moon andMizutani 2002). The tritrable acidity is higher at the mid-section than at the stem and stylar ends (Moon andMizutani 2002). However, according to Guthrie et al.(2005a), mandarin fruit are more homogenous than orangeand grapefruit.

In other fruit types, chemical composition has beenreported to vary from stem end to blossom end, and fromsun to shade sides of the fruit (Guthrie and Walsh, 1997). Inbig fruit such as pineapple and melons, the variation ismore pronounced. For example, the SSC content of thebottom of the pineapple fruit is consistently about 3°Brixhigher than the top of the fruit (Smith 1984). A very similartrend was reported by Guthrie and Walsh (1997). Theseinvestigators further reported that in addition to SSCgradient from top to bottom, the ‘sunny’ side of the fruitwas generally 1°Brix higher than the ‘shade’ side. Theseauthors suggested that in order to minimize the effect ofvariation within fruit on the obtained spectral data, NIRSassessment could be made at either a set location on thefruit or repeat the spectral acquisition at several positionsaround the whole fruit. Likewise, Guthrie et al. (2005a, b)recommended that optical and reference sampling for allcitrus fruit should be at any position around the equatorialposition of the fruit in order to best represent the entirefruit. However, this is insufficient when, for example, skindisorders are to be detected in a practical application(Nicolaï et al. 2007a).

Chemometrics

The development of an NIR spectrometric technique forassessing a quality trait in fruit or vegetable relies uponcollecting spectra of the product samples and developingcalibration equations to relate this spectral data with thequality traits ascertained using standard destructive labora-tory method (Peiris et al. 1999). One of the main problemsencountered in product characterisation from NIR spectrahas been the large volume of primary data (typicallybetween 250 and 2,000 points per spectrum, depending onthe spectrum used) (Kim et al. 2000). After the door hadopened and the potentialities demonstrated by the pioneers,NIRS encountered fast development impelled mainly byinstrumental (spectrophotometer) improvements associatedwith spectral data acquisition and their treatment (micro-computer) and which, to a great extent depended on thenew discipline of chemometrics which supplies the tools for

gathering information (Wetzel 1983; Wang and Paliwal2007). Basically, the two techniques are closely related, asNIRS would never have reached its present stage ofdevelopment without chemometrics and NIRS results arefrequently used to illustrate the power of new chemometricalgorithms (Blanco and Villarroya 2002).

The interpretation of the spectra may take place viadirect spectrum identification using data bases or increas-ingly by applying procedures of high speed chemometrics(Butz et al. 2005). Chemometrics is the study of statisticaland mathematical methods applied to the field of chemistry(McClure 2003). Calibration methods are developed usingvarious chemometric tools, such as partial least squaresregression (PLS or PLSR), multivariate linear regression(MLR), principal component regression (PCR) also referredto as principal component analysis (PCA), artificial neuralnetworks (ANN), and kernel based methods, to mention afew (Blanco and Villarroya 2002; Nicolaï et al. 2007b).

MLR is the oldest method of calibration developed in theNIRS (Norris and Hart 1965). Since its introductionnumerous other methods of calibration have been proposedincluding Fourier transform (FT) regression, PCR, PLS etc.For the NIR non-destructive prediction of fruit compositionalattributes, MLR or any of the abovementioned analysisprocedures are used to relate specific spectral regions tochanges in the concentration of a known attribute (Kim et al.2000). Simple linear multivariate least squares regressionmodels typically do not perform well because of the oftenhigh collinearity of the spectra (Saiz-Abajo et al. 2005). Forthe modelling of complicated data sets in terms of chains ofmatrices and to overcome the problem of collinearityencountered with linear multivariate least squares regressionmodels, PLS was introduced by Herman Wold in 1975(Wold et al. 2001). Guthrie et al. (2005a) compared TSSmodels of mandarin fruit developed from two regressionmethods, MLR and PLS. The authors found that the PLSmodels were better than MRL models in both calibrationdevelopment and validation on independent data sets.

PLS has been used in the development of manyprediction models for fruits and vegetables quality param-eters, and many applications are reported on citrus(McGlone and Kawano 1998; Carlini et al. 2000). Asstated by Saiz-Abajo et al. (2005), PLS is today probablythe most widely applied method in chemometrics. In‘Satsuma’ mandarin fruit, the PLS approach was used toevaluate the use of vis–NIR in measuring the qualitycharacteristics (Gómez et al. 2006). Using PCR and PLS tobuild models, the authors reported the calibration techniqueto influence the results in a similar fashion that data pre-processing affect the performance of the model. Theyreported that PLS models were better than PCR models.Besides resulting in better prediction models, PLS analysisresults in models which always have the lowest number of


latent varables (LV) since PLS models exclude LVs that arenot important to describe the variance of the qualityparameter (De Jong 1993). For instance, in Gómez et al.(2006) the calibration model for SSC content using PLShad four LV and the following statistics: r=0.97, SEC=0.160°Brix while that of PCR had six LV, r=0.969 andSEC=0.163°Brix.

According to Nicolaï et al. (2007b) the principles of thePCR and PLS methods are different. The PLS method iscommonly used in quantitative spectroscopy to correlatespectroscopic data (X) with related physico-chemical data(Y) (Saiz-Abajo et al. 2005). The method is based on LVlike PCA and PCR. Nicolaï et al. (2007a) stated that “unlikePCR in which the LVs are selected so as to explain as muchvariance of X as possible with only a limited LV, in PLS theLVs are selected in such a way that the covariance betweenX and Y is maximal”. Hence, the variation in X directlycorrelating with Y is extracted. Basically, PLS as describedby Cen et al. (2006; 2007), is a bilinear modelling methodin which the original independent information (X variable)is projected into a small number of LVs to simplify therelationship between X and Y for prediction. The authorsfurther elaborate that, ‘the Y variable is actively used inassessing the LVs to ensure that the first one is mostrelevant for predicting the Y variable’.

Different researchers have explored different regressiontechniques on citrus fruit. Using FT-NIR on intact citrusfruit to predict SSC, Lu et al. (2006) reported that PLSmodel produced better prediction results than PCR. In thework of Cayuela (2008), portable vis–NIR in reflectancewas used to predict SSC, TA and pH of ‘Valencia’ oranges.The observed spectra were similar to those described byGómez et al. (2006) in ‘Satsuma’ mandarins. However,SSC models by these authors were superior to that ofCayuela (2008) who obtained an r value of 0.95 andRMSEP of 0.51°Brix. The latter researchers used PLS andmodified PLS (MPLS) to develop models for predictingSSC, pH and TA content of mandarin fruit and reportedPLS calibration models for SSC to perform better thanMPLS models. Due to very low coefficients of calibrationobtained for TA and pH, none of the calibration methoddelivered acceptable prediction model for these parameters.However, a comparison between mandarin and oranges isnot recommended for oranges are larger in size and havedifferent structure to mandarin.

Recently, a modification to PLS called iPLS whichsearches for spectral interval that is particularly informativewith respect to the parameter under consideration has beendeveloped to simplify the model and improve its predictiveability (Nørgaard et al. 2000). The main advantage of theiPLS method is to present a local regression model in agraphical display, to choose better interval and permit acomparison between interval and the full spectrum models.

In addition, wavelength selection enhances the stabilityof the calibration model resulting from the collinearityin multivariate spectra. Furthermore, wavelength selec-tion helps to interpret the relationship between themodel and the sample compositions. In a study aimedat investigating the feasibility of charge-coupled deviceNIRS in the wavelength range of 600–980 nm, Liu etal. (2010b) simplified and yet improved the predictiveability of SSC model with r value of 0.92 and RSMEP of0.65°Brix. In general, PLS has been proved by manyresearchers to be a better method than the others, such asMLR and PCR.

PLS method is based on linear models and, consequently,unsatisfactory results may be produced when non-linearity ispresent (Liu et al. 2010a, b). ANNs are commonly used formodelling non-linearity. Some of the advantages of usingANNs include the flexible learning algorithm, diversenetwork topology, fast learning algorithm and high errortolerance. Recently, the authors compared the PLS toPCA coupled with back propagation neural network(BPNN) (PCA-BNN) to develop the model for predictingSSC of intact navel orange fruit. In contrast to resultsobtained by Lu et al. (2006) comparing PCA and PLS,reported PCA-BPNN method to give better predictionmodel than PLS. Alternative nonlinear techniques includekernel based models such as kernel PLS and least squares–support vector machines. This framework allows the PLScalibration to be carried out in a space of nonlinearilytransformed input data—the so-called feature space—without actually carrying out the transformation. Theadvantage is that the feature space has a potentially muchricher structure than the space spanned by the originalinput data set. However, Nicolaï et al. (2006) found thatkernel PLS yields comparable results to those obtainedwith ordinary PLS (covariance kernel), irrespective of thekernels which had been evaluated. It is questionnablewhether any improvement of accuracy of the calibrationmodels would be worthwhile given the noise due tobiological variability and model robustness issues.

Spectral Pre-processing

The spectra of solid and scattering samples such as intactfruit and vegetables are influenced by physical propertiessuch as shape, size etc. (Leonardi and Burns 1999). Thiscreates noise problems when analysing quality parametersfor which such physical characteristics are not important(Blanco and Villarroya 2002). In the analysis of spectra,there are large baseline shifts and noises in the spectra withbroad wavelength regions, thus the selection of suitable pre-processing or pre-treatment method is an important step inthe process of spectral analysis (Cen et al. 2006). In order tofacilitate handling and to develop more simple and robust


models, the complex spectral data are often pretreated bystatistical procedure (Butz et al. 2005).

Some of the more frequently used pre-treatment methodsfor NIR spectra include normalisation, derivatives (usuallyfirst and second), the multiplicative signal correction alsoknown as multiplicative scatter correction (MSC), thestandard vector normalisation (SNV), smoothing usingmoving average and detrending (Blanco and Villarroya2002). Derivative spectroscopy is a method which hasdemonstrated success in correcting light scattering contri-butions provided that the spectrum describes a largewavelength range (Leonardi and Burns, 1999). To reducethe fruit size effect on citrus spectra, Kawano et al. (1993)normalised the second derivative spectra by dividing it bythe second derivative value at 844 nm, which had a highcorrelation to the diameter of the fruit. However, derivativespectroscopy does not correct the pathlength variations.MSC is suggested to correct for pathlength variations.

In Gómez et al. (2006), two types of pre-processing wereemployed. First was smoothing using moving averagemethod at different data points and the second was theuse of MSC. The later technique was used to correct foradditive and multiplicative effects of spectra. From thisstudy it was observed that pre-processing with MSCresulted in models that were better than models withoutMSC. Working on citrus with NIR in reflectance mode, Luet al. (2006) used three types of data pretreatments:constant, SNV and MSC. They obtained similar statisticalparameters for all three pre-treatment methods and arguedthat different spectral pre-treatment is not useful forimproving quality of models. In contrast, when similarpre-treatment techniques, SNV and MSC, were applied tothe spectra of ‘Navel’ oranges by Liu et al. (2010a), theMSC spectral pre-treatment method showed superior modelthan SNV.

Model Accuracy

The accuracy of NIRS models for fruit quality prediction isusually described by the value of the R (Eq. 1), the rootmean square error of calibration (RMSEC) (Eq. 2) and theRMSECV or RMSEP (Eq. 3) (Lu et al. 2006, Sun et al.2009, Liu et al. 2010a, b, Bobelyn et al. 2010; Camps andChristen 2009).

A good model should have the lower RMSEC,RMSECV and RMSEP, higher correlation coefficient (r orR) or coefficient of determination (r2 or R2), whichrepresent the proportion of explained variance of theresponse variable in the calibration (Rc

2) or validationdataset (Rv

2). Other statistical parameters explaining a goodmodel are low average difference between predicted andmeasured values (Bias) (equation 4) and a small differencebetween RMSEC and RMSEP. In addition, a good model

should have as little as possible number of LV or principalcomponents.

R ¼ 1�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

P

ycal � yactð Þ2P

ycal � ymeanð Þ2

s

ð1Þ

RMSEC ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

X

ðycal � yactÞ2=nq

ð2Þ

RMSECV or RMSEP ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

X

ðypred � yactÞ2=nq

ð3Þ

Bias ¼ 1

n

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

X

ðypred � yactÞ2q

ð4Þ

Where:

n The number of spectrayact The actual valueymean The mean valueycal The calculated valueypred The predicted value of the fruit attribute.

Model Robustness

The steps followed in the development of a calibrationequation are most of the time standard, but the conclusionsand prediction models obtained are not universally appli-cable to all fresh fruit or that fruit type because of the effectof fruit surface characteristics and internal chemistry on theNIR reflectance spectra. The application of NIR spectros-copy to a given fruit commodity requires an assessment ofthe robustness of the developed calibration model acrosspopulations of fruit grown under differing conditions(Guthrie et al. 2005b). A robust model was defined byNicolaï et al. (2007a) as a model with prediction accuracythat is relatively insensitive towards unknown changes inexternal factors. These authors went further to list thewithin-tree variability (tree age, crop load, fruit positionwithin the tree and light effect), within-orchard variability(location of the tree and light effects), orchard variability,fruit age and seasonal variability as main factors which mayaffect model performance.

In addition, for a calibration model to be called robust, itshould be able to perform with similar accuracy whentransported from one instrument to another. Guthrie et al.(2005b) stated that differing growing conditions may resultin differences in physical and chemical properties of thefruit. This change in fruit physico-chemical properties resultin altered fruit optical characteristics and spectral bandassignment (Golic et al. 2003; Guthrie et al. 2005b). Region


and seasonal variation has been considered important toinclude in the NIR experimental design as differentgrowing conditions in different places and seasons couldaffect fruit analyte levels with concomitant effects NIRpredictions McGlone et al. (2002). In support of this notion,Nicolaï et al. (2007a) further stated that external validationis of prime importance for the successful application ofmultivariate calibration models.

While model robustness has been investigated for otherspecies (Lovász et al. 1994; Guthrie et al. 1998; McGloneand Kawano 1998; Peirs et al. 2002; Guthrie et al. 2006;Golic and Walsh 2006; Bobelyn et al. 2010), very littleliterature reporting on citrus models obtained from morethan one harvest population has been documented (Miyamotoand Kitano 1995; Ou et al. 1997; Guthrie et al. 2005b). Thelatter researchers reported on the robustness of NIRmodels for the evaluation of attributes related to eatingquality of intact mandarins. Model performance wasvalidated in terms of fruit from different harvest daysfrom a single tree, different harvest localities and differentseasons. They found that model predictions were morevariable across seasons than across harvest days andlocation. However, previous studies indicated that modelperformance should be more stable across seasons, for agiven variety, than across varieties, in a given season(Peiris et al. 1998a, b; Miyamoto and Kitano 1995). Mostdocumented literature did not involve testing the calibrationmodels against fruit from independent populations such as,different harvest dates, localities, temperatures, cultivar,size, shape etc. Therefore, NIR model developed for aparticular cultivar, for instance, is limited to the calibrationcultivar. Skin thickness, pulp texture, fruit shape, size andcomposition are major factors that differ between citruscultivars (Cayuela 2008). Therefore, histological differ-ences may influence measurements of NIR of citrus fruit.One of the weaknesses of NIRS is that its application inquality evaluation for different fruit species needs to bedesigned according to the fruit size, the thickness of its skinand the specific attributes to be tested. According to Louwand Theron (2010), the best way to develop robust modelsfor biological products is to acquire calibration data over asufficient period of time to obtain an appropriate range ofinstrumental and environmental conditions.

Temperature influences are often described as adrawback of NIRS measurements (Peirs et al. 2003;Zude et al. 2008). Performance of calibration model isaffected by sample temperature primarily through thestrong effect of temperature on H bonding and thus onthe absorption bands related to OH (Golic et al. 2003).Calibration models developed across a wide range oftemperatures (5.5–23.5°C) are expected to be more robustin terms of predicting analyte levels of samples at a rangeof temperatures (Peirs et al. 2003; Zude et al. 2008).

However, the accuracy of the model is potentiallydiminished (Guthrie et al. 2005b). Results obtained inthese studies were in agreement with those obtained byMiyamoto and Kitano (1995) and Kawano et al. (1993),who recommended that samples scanned at a range oftemperatures, should be included in the calibrationpopulation in order for the model to be robust in predictionof samples varying in sample temperature. Given thesensitivity of water bands to temperature, DM calibrationmodels have been previously reported to be more sensitiveto temperature than TSS models (Golic et al. 2003).

Commercial Adoption of NIRS Technology

The important objective governing technology developmentresearch in postharvest science is to explore the possibilitiesof the technology for in-line assessment of quality. Whilescientific literature is replete with calibration and predictionstatistics for the use of NIRS for the non-destructiveassessment of fruit quality and the technology is incommercial use for a number of commodities includingcitrus fruit (Golic and Walsh 2006), NIR technology forwide application in non-destructive measurement of fruitquality is still in development. A review by Walsh (2005)discussed the application and limitations to adoption ofcommercially available, low cost, miniaturised NIR spec-trophotometers for the assessment of the sugar content ofintact fruit, including citrus. Commercial application ofNIRS to fruit sorting was first initiated in Japan in 1990,and has been applied to pack-house fruit sorting lines forsweetness of citrus and other fruit since the mid 1990s inJapan, and more recently in other countries. Although thetechnology has been available for commercial applicationfor over a decade, adoption by the packing industry hasbeen low outside Japan. Similar to most technologies in theagricultural sector, the limitations to NIRS adoption bycommercial packing industry could be attributed to cost,technical limitations, grower resistance and supply chainlimitations. However, the technology is sufficiently welldeveloped, and offered by a sufficient number of manu-facturers to overcome ‘technical’ limitations. It is likely thattechnology adoption will be rapid once consumers are ableto recognize, and consistently source, fruit of superiorquality. Although the technology is in commercial use for anumber of commodities, commercial operators have notpublished such information.

Specific Applications of NIRS on Citrus Fruit

Measurement of Internal Quality Attributes

Citrus quality parameters such as SSC and TA are related toorganic molecules that contain bonds C–H, O–H, C–O and


C–C; hence, it is possible to use NIR methods to determinetheir concentration. Tables 2, 3 and 4 summarise theapplications of NIRS to assess internal quality parametersof citrus fruit. These literature evidence shows that someinternal properties are more accurately measured with NIRthan others. For instance, the NIR technique has signifi-cantly greater accuracy for determining SSC in citrus thanany other quality parameters such TA, and the low successcould be attributed to the fact that organic acidsconcentration on intact fruit is relatively low; hence,calibration of this attribute is likely to represent secondarycorrelations to attributes related to fruit maturity (Guthrieet al. 2005a).

Measurement of External Quality Attributes

Appearance is the primary parameter used to evaluatequality of fruit and vegetables and the presence of skindefects is one of the most influential factors in the price offruit. Under present fruit grading systems, fruit with slightexternal defects are graded and marketed with sound fruit,thereby reducing the quality of the batch. Alternatively,fruit with slight defects are graded and removed togetherwith seriously damaged fruit, thus causing economic loses(Blasco et al. 2007a). The challenge is significant regardingcitrus rind disorders that do not manifest during harvestgrading and postharvest treatments but develop about 1–5 weeks after harvest; such as rind breakdown disorder(RBD) of ‘Nules Clementine’ mandarins (Citrus reticulateBlanco.) (Cronje 2009), RBD of ‘Navel’ orange (Citrussinensis L. (Osb.)) (Alferéz et al. 2003) and non-chillingpostharvest rind pitting of ‘Marsh’ grapefruit (Citrusparadisi Macf.) (Alferéz and Burns 2004). The challengeis therefore to develop non-destructive technology todetermine rind quality in the packing line to assist insorting and segregation of fruit into quality grades. There istherefore a need to develop an objective, fast and non-destructive assessment that can be used to determine/predictcitrus fruit susceptibility to rind disorders accurately. Non-visible information, such as that provided by NIR region ofthe spectrum can improve the inspection by detectingspecific defects or allowing the detection of non-visibledamages (Blasco et al. 2007b)

Most current non-destructive quality measurement tech-nologies have been developed to assess fresh fruit accord-ing to their internal quality attributes (Butz et al. 2005).Very limited research work has been conducted to developa technology that can assess, predict and monitor thephysiological disorders and rind physiological disorders ofcitrus fruit in particular. Nevertheless, NIR has been usedsuccessfully to detect surface bruising in apple (Geeola etal. 1994), surface defects in peach (Miller and Delwiche1991), storage disorders in kiwifruit (Clark et al. 2004) and

drying internal disorder in Tangerine citrus (Peiris et al.1998c). The trend has constantly shifted towards develop-ing reliable and cost-effective technologies to non-destructively screen fruit physiological disorders. Zheng etal. (2010) used NIR in the reflectance mode to predictoleocellosis sensitivity in citrus fruit. Meglinski et al.(2010) established that internal and external quality defectsof onions can be effectively observed and screened byoptical coherence tomography (OCT) system, and this non-invasive method is suitable for examining intact plants.OCT produces two-dimensional (2D) images of planttissues at a penetration depth of 1–2 mm from the surface(Sapozhnikova et al. 2003). Detailed aspects of OCT theoryand applications have been discussed in comprehensivereviews by Tomlins and Wang (2005) and Fercher et al.(2003). NIRS and similar emerging non-destructive tech-nologies such as OCT offer potential for the non-invasiveassessment of the chemical composition and changesthereof, such as carbohydrate content in the flavedo ofintact fruit that might predispose fruit to the development ofrind disorders.

Multispectral and Hyperspectral Imaging

In recent years, extensive research has been conducted todevelop techniques for detecting defects on fruit andvegetables (Xing et al. 2005). Systems based on varioustypes of electromagnetic methods have been investigated.The analytical spectral regions of these systems include theultraviolet, vis–NIR, mid-infrared and far-infrared regions.Imaging techniques have been developed and successfullyapplied as inspection tools for quality assessment of avariety of fruits (Xing et al. 2003, 2005; Mehl et al. 2004).Multispectral imaging involves making images using morethan one spectral component of the electromagnetic energyfrom the same region of an object and at the same scale.Hyperspectral imaging on the other hand, integratesconventional imaging and spectroscopy to attain bothspatial and spectral information from an object. Hyper-spectral images are made up of hundreds of contiguouswavebands for each spatial position of a sample studied andeach pixel in an image contains the spectrum for thatspecific position (Gómez-Sanchis et al. 2008). Basically,the development of the multispectral and hyperspectralimaging provides the possibility to capture images not onlyin the visible range but also extend to capture multipleimages at different wavelengths in the vis–NIR regions(Mehl et al. 2004; Nicolaï et al. 2006; Qin et al. 2008).These analytical technologies, which combine vis–NIRSwith digital imaging, now allow both spatial and spectralinformation to be obtained simultaneously. Further infor-mation on the principles of these technologies could befound in a review by Gowen et al. (2007)


Tab

le2

Overview

ofapplications

ofNIR

spectroscopy

tomeasure

fruitqu

ality

ofMandarinfruit

Cultiv

arParam

eter

Mod

eWavelength(nm)

Preprocess

Validation

Analysis

R/R

2Predictionerror

Reference

Imperial

SSC

Interactance

306–

1,13

0SVN

Cross-validation

MLR

R2=0.75

RMSEC=0.4

Guthrie

etal.(200

5a;b)

DM

Detrend

ing

MPLS

R2=0.90

RMSEC=0.6

TA1stder.

R2=0.30

RMSEC=0.2

%Juice

2ndder.

R2=0.20

RMSEC=5.0

Nanfeng

SSC

Transmittance

400–

1,04

0Smoo

thing

–BPNN

PLS

R=0.93

RMSEP=0.65

Sun

etal.(200

9)TA

MSC

R=0.66

RMSEP=0.09

Vit.

C1stder.

R=0.81

RMSEP=2.7

Colou

r2n

dder.

R=0.57

RMSEP=0.81

Nafeng

SSC

Transmittance

350–

1,04

0Cross

andexternal

iPLS

R=0.92

RMSEP=0.65

Liu

etal.(201

0b)

TAR=0.64

RMSEP=0.09

Pagetang

elo

SSC

400–

1,00

0Fullcross-valid

ation

PLS

R=0.85

RMSECV=0.45

Anton

ucci

etal.(201

0)TA

R=0.88

RMSECV=0.09

Sastuma

SSC

Reflectance

400–

1,00

0Savitzky

Fullcross-valid

ation

PLS

R=0.84

RMSECV=0.59

Anton

ucci

etal.(201

0)TA

Golay

R=0.81

RMSECV=0.07

Satsuma

SSC

Reflectance,interactance

andtransm

ittance

500–

1,10

02n

dder.SNV

Cross

andexternal

PLS

R2=0.93

RMSEP=0.32

McG

lone

etal

(200

3b)

TAR2=0.65

RMSEP=0.15

Satsuma

SSC

Reflectance

350–

2,50

0Smoo

thing,

MSC

–PLSandPCA

R=0.94

RMSEP=0.33

Góm

ezet

al.(200

6)pH

R=0.80

RMSEP=0.18

Firm

R=0.83

RMSEP=8.53

Satsuma

SSC

Transmittance

800–

1,10

0Normalisation

––

––

Tsuchikaw

aet

al.(200

3)TA

Satsuma

SSC

Transmittance

680–

1,23

52n

dder.

–MLR

R=0.99

SEP=0.32

Kaw

anoet

al.(199

3)


Tab

le3

Overview

ofapplications

ofNIR

spectroscopy

tomeasure

fruitqu

ality

oforange

fruit

Cultiv

arParam

eter

Mod

eWavelength

(nm)

Preprocess

Validation

Analysis

R/R

2Prediction

error

Reference

Gannan

SSC

Reflectance

800–

2,50

0SNV

–PLS

R=0.99

5RMSEP=0.75

Luet

al.(200

6)MSC

PCR

Navel

SSC

Reflectance

350–

1,80

0MSCandSNV

Cross-andexternal

valid

ation

PLSand

PCA-BPNN

R=0.90

RMSEP=0.68

Liu

etal.(201

0a)

Navel

Vitamin

CReflectance

800–

2,50

0COE,SLS,SVN,

MMN,MSC,1st

derand2n

dder

Cross-validation

PLS

R=0.96

RMSECV=3.9

Xia

etal.(200

7)

Sangu

inelli,

Valencia,

Salustiana,Navelate

SSC

Reflectance

350–

2,50

0MN

Externalvalid

ation

PLS

R=0.92

RMSEP=0.74

Cayuela

and

Weiland

(201

0)pH

MSC

R=0.90

RMSEP=0.15

TAR=0.86

RMSEP=0.17

MI

R=0.85

RMSEP=1.92

FW

R=0.97

RMSEP=16

.52

FR=0.76

RMSEP=1.05

JVR=0.92

RMSEP=7.05

JV/FW

R=0.89

RMSEP=0.04

RW

R=0.97

RMSEP=12

.98

FCI

R=0.90

RMSEP=2.65

JCI

R=0.86

RMSEP=66

.78

ValenciaLate

SSC

Reflectance

578–

1,85

0SVN

anddetrending

Internal

andexternal

valid

ation

PLSandMPLS

R2=0.91

RMSEP=0.51

Cayuela

(200

8)TA

R2=0.56

RMSEP=0.33

pHR2=0.45

RMSEP=0.49

Trovita

Oleocellosisrate

Different

reflectance

325–

1,07

5Mov

ingaverageMSC

Cross-validation

PLS

R=0.98

SEP=0.00

79Zheng

etal.(201

0)Oleocellosisdegree

R=0.99

SEP=0.00

56


Tab

le4

Overview

ofapplications

ofNIR

spectroscopy

inothercitrus

fruitandcitrus

prod

ucts

App

lication

Param

eter

Mod

eWavelength

(nm)

Preprocess

Validation

Analysis

R/R

2Prediction

error

Reference

22varieties

Fructose

Transmittance

1,10

0–2,50

01stder.

Cross-validation

PLS

R2=0.95

–Tew

ariet

al.(200

8)Glucose

2ndder.SVN

R2=0.94

Sucrose

R2=0.98

Grapefruit

Lim

onene

Reflectance

1,10

0–2,50

0–

Cross-validation

PLS

R2=0.99

SECV=0.65

Steueret

al.(200

1)Myrcene

Transmittance

R2=0.87

SECV=0.06

α-Pinene

R2=0.98

SECV=0.10

β-Pinene

R2=1.00

SECV=0.28

Sabinene

R2=0.95

SECV=0.15

γ-Terpinene

R2=0.99

SECV=0.25

Optical

rotatio

nR2=0.99

SECV=1.18

Aldehyd

eR2=0.79

SECV=0.33

Indian

River

red

SSC

Reflectance

400–1,10

0–

–PLS,NN

R<0.7

–Miller

andZud

e(200

2)

Different

brands

oforange

juice

SSC

Reflectance

325–1,07

5SVN

Savitzky

–Golay

Fullcross-valid

ation

PLS

R=0.98

RMSEP=0.73

Cen

etal.(200

6)pH

R=0.96

RMSEP=0.06

Different

brands

oforange

juice

Citric

acid

Reflectance

325–1,07

5MSC

Cross-validation

PLS

R=0.94

RMSEP=0.59

6Cen

etal.(200

6)Tartaricacid

R=0.93

RMSEP=0.01

3

Trovita

Oleo.

rate

Different

reflectance

325–1,07

5Mov

ingaverageMSC

Cross-validation

PLS

R=0.98

SEP=0.00

79Zheng

etal.(201

0)Oleo.

degree

R=0.99

SEP=0.00

56

Tarocco

citrus

leaves

NTransmittance

400–1,00

0Baseline,

MSC

Cross-validation

PLS

R=0.91

SEP=0.04

9Menesattiet

al.(201

0)P

R=0.43

SEP=0.01

8

KR=0.99

SEP=0.05

8

CA

R=0.95

SEP=0.30

4

Mg

R=0.94

SEP=0.04

8

Fe

R=0.92

SEP=6.05

4

Mn

R=0.93

SEP=1.63

7

Zn

R=0.89

SEP=0.97

2


Several spectroscopic studies demonstrated the possibil-ity of using multispectral and hyperspectral imagingtechniques for identifying external skin damages in citrusfruit (e.g. Blasco et al. 2007a; Qin et al. 2008). A recentreview on advances in machine vision applications forautomatic inspection and the application of hyperspectraland multispectral imaging techniques was given by Cuberoet al. (2010). Tables 5 and 6 present an overview ofapplications of multispectral and hyperspectral imagingsystems to assess external quality parameters of citrus fruit,including different peel disorders, such as, citrus canker(Balasundaram et al. 2009), anthracnose, stem-end injury,green mould, oleocellosis and chilling injury (Blasco et al.2007a) and different common peel defects (Blasco et al.2007b). Results by Blasco et al. (2007a, b) showed that thecontribution of non-visible information detected in the NIRregion can improve the detection and identification of somedefects. Blasco et al. (2009) showed that the system couldsuccessfully discriminate between 11 defects in 86% ofcases. Using only visible information, the results were alsoacceptable (82%) but the confusion between serious andslight defects increases since green mould could be detectedusing UV in 97% of cases and anthracnose was detectedusing NIR in 95% of cases. Although very limited work hasbeen reported on citrus rind disorders, spectrophotometricsurface detection offers an alternative or complementarymethod to machine vision in fruit and vegetable research(Xing et al. 2003). The vast accumulation of data withmultispectral imaging creates novel opportunities andapplications. However, there are still many unexploredapplications to be investigated. For instance, reducing theprocessing time is still a challenge since it is too high toallow both the acquisition and image processing software tobe implemented in a real-time system. However, furtheradvances in technology can help this to be achieved.

Future Prospects

NIR technology is one of the most advanced non-destructive quality evaluation systems available for a widerange of applications. There are considerable researchactivities into the investigation of NIR-based sensors fordetermining various quality factors in a range of fruit andvegetables, including citrus. The move to tomographicimaging would seem the next logical step as prercusor tothe development of true three-dimensional (3D) imaging ofcitrus and other fresh food products (Hebden et al. 2002).Already, numerous attempts have been undertaken toimplement non-destructive techniques of acquiring imagesof plant tissues (Lammertyn et al. 2003a, b; Kemsley et al.2008). Due to the poor transparency of biological tissues inthe vis–NIR region, optical methods were previouslyregarded inapplicable for the visualisation of structural T

able

5Sum

maryof

applications

ofmultispectralim

agingsystem

toidentifyexternal

defectsanddiseases

ofcitrus

fruit

Com

modity

Quality

parameter

Imagingsystem

Wavelength

Dataprocessing

Optim

alwavelength

Accuracy

Reference

Valencia

Defects

CCD

(1,392

×1,04

0pixels)

Vis/NIR

(550–800

nm)

PCA,LD

andANNs

650nm

93.20%

Bulanon

etal

(201

0)

Clemenules

Marisol,

Fortune

andValencia

Defects

3-CCD

(768

×57

6pixels

and0.17

mm/pixel

resolutio

n)–

PCA

andMIA

–93

.30%

Lop

ez-G

arciaet

al.(201

0)–

–93

.30%

––

96.60%

––

93.30%

Grapefruit

Canker

Vis/NIR

and3-CCD

RGB

(640

×32

0pixels)

200–2,500nm

DA

500–80

0nm

100%

Balasundaram

etal.(200

9)

Mandarinandorange

Defects

Vis/NIR

andmon

ochrom

atic

RGB

camera(768–576

pixelsand

0.17

mm/pixel

resolutio

n)

RGB(350–400

nm)

LDA

–86

%Blascoet

al.(200

9)Vis/NIR

(400–1,800

nm

Mandarinandorange

Defects

Vis/NIR,UV,Fl(3-CCD

768×57

6pixelsand

.017

mm/pixel

resolutio

n)

RGB(350–400

nm)

DA

–95

%Blascoet

al.,20

07b

Vis/NIR

(400–1,800

nm)

Lem

ons,mandarins

andoranges

Defects

Vis/NIR

and2-CCD

NIR

and

RGB(768

×57

6pixelsand

0.17

mm/pixel

resolutio

n)

Centred

at75

0nm

––

94%

Aleixos

etal.(200

2)–

–93

%

––

94%

CCD

charge-cou

pled

device,DAdiscriminateanalysis,UVultraviolet,LDAlin

eardiscriminateanalysis,PCAprincipalcompo

nent

analysis,ANNsartificialneural

networks,RGBred,

greenand

blue

colour

camera,

MIA

multip

leim

ageanalysis


features of such tissues. As stated earlier in this review andby Sapozhnikova et al. (2003) that although biologicaltissues are optically turbid media, they do not stronglyabsorb radiation in the NIR region, 0.83 μm≤λ≤1.3 μm inparticular. In addition, biological tissues exhibit sufficientlystrong scattering properties.

Optical tomography (OT) is one such non-invasiveanalytical technique currently available, at least to researchers,that is suitable for examining internal structures of plantand animal tissues. OT reconstructs 3D images of objectsbased on transmission or scattering images. Since its firstinception in the biological field, OT has been utilizing twofundamental techniques, which are optical diffuse tomog-raphy and OCT. Although a wide range of tomographicimaging techniques and sensors has been explored, basedon merits, OCT has been advocated to be suitable forexamining internal structures of biological tissues (Fercheret al. 2003; Huang et al. 1991; Lu et al. 2004). OCTprovides a means to visualise the internal structure ofbiological objects with high resolution in real-time mode(Meglinski et al. 2010). The image acquisition time 1–3 sper tomogram. The method is based on the interferometricmeasurements of NIR light back-scattering from the fruit(Sapozhnikova et al. 2003). The radiation in the NIR regionused in OCT systems is suitable and safe for examiningintact plants for its non-ionizing characteristics (Fercher etal. 2003; Kemsley et al. 2008). In addition to its non-destructive properties, other advantages of OCT includehigh depth in scattering media such as fruit, provides themeans to observe and monitor physiological activity andchanges of plant tissues during growth and development.Other advantages include the fact that depth resolution isdecoupled from transverse resolution. OCT images aresynthesized from a series of adjacent interferometric depth-scans performed by a straight propagation low-coherenceprobing beam. The main parameters of an OCT instrumentare its optical power and wavelength, its penetration depth,resolution and sensitivity, and its image acquisition rate.The choice of wavelength is highly dependent on the natureof the sample. An important feature of an OCT system isthe weakest sample reflectivity yielding to a signal powerequal to noise of the system (Fercher et al. 2003).

Recently, Kutis et al. (2005) conducted a comparativeanalysis of two optical methods, namely, OCT and opticalcoherence microscopy (OCM) to visualise plant tissues.The latter method combines the advantages of the confocalmicroscope and OCT, the two reputed instruments toobserve objects with strong light scattering properties. Theauthors reported a higher spatial resolution of the OCMmethod compared with the OCT method. Although OCMprovided higher spatial resolution and appeared morereliable that OCT in identification of plant morphologicalstructures, it should be noted that both methods have theirT

able

6Sum

maryof

applications

ofhy

perspectralim

agingsystem

toidentifyexternal

defectsanddiseases

ofcitrus

fruit

Com

modity

Qualityparameter

Imagingsystem

Wavelength

Imagepre-processing

Data

processing

Optim

alwavelength

Accuracy

Reference

Mandarins

Green

mou

ld(Penicillium

digitatum)

Mon

ochrom

ecamera(551

×55

1pixels

and3.75

mm/pixel

resolutio

n)Vis/NIR

(320–1,

100nm

)Fastergeom

etrical

correctio

nCA

andLDA

–91

.0%

Gom

ez-Sanchis

etal.,20

08

Mandarinandorange

Defects

RGBCCD

(768

×57

6and

.017

mm/pixel

resolutio

n)–

Smoothingusing

peer

grou

pfiltering

––

95.0%

Blascoet

al.(200

7a)

Rubyredgrapefruit

Canker

EMCCD

(658–496

pixels)

400–900nm

Flat-fieldcorrectio

nPCA

553,

677,

718

and85

8nm

92.7%

Qin

etal.(200

8)

Rubyredgrapefruit

Canker

EMCCD

(1,004

×1,002pixels)

450–930nm

Flat-fieldcorrectio

nCA,DA

andSID

–95

.7%

Zhaoet

al.(201

0)

Rubyredgrapefruit

Canker

EMCCD

(1,004

×1,002pixels)

450–930nm

Flat-fieldcorrectio

nSID

–96

.2%

Qin

etal.(200

9)

CAcorrelationanalysis,CCD

charge-cou

pled

device,DAdiscriminateanalysis,EMCCD

electron

multip

lyingcharge-cou

pled

device,LDAlin

eardiscriminateanalysis,PCAprincipalcompo

nent

analysis,SID

spectral

inform

ationdivergence,RGBred,

greenandblue

colour

camera


advantages and drawbacks. In retrospect, both OCM andOCT as non-destructive techniques could open newperspective in plant physiology and morphology studies.

Because of the relatively low value of horticulturalproducts compared to biomedical and pharmaceuticalproducts, the horticultural industry had been users insteadof developers of advanced technologies including sensors.Similarly, the OCT method has been previously usedexclusively by the biomedical research for differentialdiagnosis of tumours (Sapozhnikova et al. 2003; Kutis etal. 2005). The feasibility of OCT for imaging of plants andrelated food materials such as fruit is currently receivingattention among researchers. Sapozhnikova et al. (2004)reported that OCT method can be used to visualise not onlyplant tissues and tissue boundaries but also the structure ofindividual cells. This non-invasive method is suitable forexamining intact plants and it is capable of producing 2Dimages of plant tissues at a penetration depth of 1–2 mmfrom the surface (Sapozhnikova et al. 2003). The techniquehas opened new avenue of research towards non-invasiveprediction and biomonitoring of peel disorders such as peelpitting that do not manifest visually at harvest but after 1–5 weeks postharvest.

Conclusions

Over the years, much technological progress has been madein the development of non-destructive quality evaluationsystems for the food industry, including fruit. Among non-destructive quality assessment techniques, NIRS is arguablythe most advanced with regard to instrumentation, applica-tions, accessories, and availability of suitable chemometricsoftware packages. In addition to monitoring and predictingexternal quality attributes of citrus fruit, NIRS also offersconsiderable potential for non-invasive measurement of thechemical composition, and changes thereof, such ascarbohydrate content, in the flavedo of intact citrus fruitthat might predispose fruit to the development of rinddisorders. Very limited research work has been reportedwhich apply NIRS technology to assess, predict andmonitor the presence of physiological disorders in the rindand flesh of citrus fruit. Recently, the trend has shiftedtowards developing reliable and cost-effective technologiesto non-destructively screen fruit physiological disorders.Innovative technologies such as OCT and OCM offer newprospects. One of the challenges facing future applicationof NIRS technology is the inherent variability in fruit andsimilar biological products. Consequently, researchers haveapplied different NIRS settings and analytical frameworksto improve prediction, but it is conceivable that the choiceof these parameters could affect both model accuracy androbustness and this deserves attention for future research.

Increasing consumer demand for consistent supply of awide range of safe nutritious and traceable food products,including citrus and other horticultural fresh produce, assuresthe need for the development of innovative non-destructivetools for field and laboratory measurement as well as in-linesorting and grading based on both external and internalproduct attributes. In addition to consumer demand forinformation on fruit internal quality attributes related to thenutritional value, citrus fruit also develop physical defects andphysiological disorders which exacerbate during postharvesthandling and marketing. NIRS and similar technological toolsreviewed in this paper offer the potential for non-destructivemeasurement and prediction of these quality problems.Equipment cost and consumer willingness to pay for freshfruit graded in terms of internal quality attributes such as sugarcontent, acidity and vitamins remains a major challenge forfuture development and commercialisation of NIRS andsimilar non-invasive tools in citrus and other fruit sectors.

Acknowledgements This project was funded by the DST/NRFSouth African Research Chairs Initiative. The authors are also gratefulto the Perishable Products Export Control Board and the South Africa/Flanders Research Cooperation Programme for financial supportwhich made it possible to undertake the study.

References

Aleixos, N., Blasco, J., Navarrón, F., & Moltó, E. (2002). Multispectralinspection of citrus in real-time using machine vision and digitalsignal processors. Computers and Electronics in Agriculture, 33,121–137.

Alferéz, F., Agustí, M., & Zacarìas, L. (2003). Postharvest rind staining in‘Navel’ oranges is aggravated by changes in storage relativehumidity: effect on respiration, ethylene production and waterpotential. Postharvest Biology and Technology, 28, 143–152.

Alferéz, F., & Burns, J. (2004). Postharvest peel pitting at non-chillingtemperatures in grapefruit is promoted by changes from low tohigh relative humidity during storage. Postharvest Biology andTechnology, 32, 79–87.

Antonucci, F., Pallottino, F., Paglia, G., Palma, A., D’Aquino, S., &Menesatti, P. (2010). Non-destructive estimation of mandarinmaturity status through portable VIS-NIR spectrophotometer. Foodand Bioprocess Technology, 3. doi:10.1007/s11947-010-0414-5.

Balasundaram, D., Burks, T. F., Bulanon, D. M., Schubert, T., & Lee,W. S. (2009). Spectral reflectance characteristics of citrus cankerand other peel condition of grapefruit. Postharvest Biology andTechnology, 51, 220–226.

Blanco, M., & Villarroya, I. (2002). NIR spectroscopy: a rapid-responseanalytical tool. Trends in Analytical Chemistry, 21, 240–250.

Blasco, J., Aleixos, N., &Moltó, E. (2007). Computer vision detection ofpeel defects in citrus by means of a region oriented segmentationalgorithm. Journal of Food Engineering, 81, 535–543.

Blasco, J., Aleixos, N., Gómez, J., & Moltó, E. (2007). Citrus sortingby identification of the most common defects using multispectralcomputer vision. Journal of Food Engineering, 83, 384–393.

Blasco, J., Aleixos, N., Gómez-Sanchís, J., & Moltó, E. (2009).Recognition and classification of external skin damage in citrusfruits using multispectral data and morphological features.Biosystems Engineering, 10, 137–145.


http://dx.doi.org/10.1007/s11947-010-0414-5

Bobelyn, E., Serban, A., Nicu, M., Lammertyn, J., Nicolaï, B. M., &Saeys, W. (2010). Postharvest quality of apple predicted by NIR-spectroscopy: study of the effect of biological variability onspectra and model performance. Postharvest Biology andTechnology, 55, 133–143.

Bulanon, D. M., Burks, T. F., & Alchanatis. (2010). A multispectralimaging analysis for enhancing citrus fruit detection. Environ-mental Control and Biology, 48(2), 81–91.

Butz, P., Hofmann, C., & Tauscher, B. (2005). Recent developmentsin non-invasive techniques for fresh fruit and vegetable internalquality analysis. Concise Reviews in Food Science, 70, 131–141.

Camps, C., & Christen, D. (2009). Non-destructive assessment ofapricot fruit quality by portable visible-near infrared spectroscopy.Food Science and Technology, 42, 1125–1131.

Carlini, P., Massantini, R., & Mencarelli, F. (1999). Wavelength selectionmethods for PLS-based vis–NIR evaluation of SSC in fresh fruits.In: Proceedings of the NIR’99, 9th International Conference onNear-Infrared Spectroscopy, Verona, Italy, 13–18 June 1999.

Carlini, P., Massantini, R., & Mencarelli, F. (2000). Vis–NIRmeasurement of soluble solids in cherry and apricot by PLSregression and wavelength selection. Journal of Agricultural andFood Chemistry, 48, 5236–5242.

Cayuela, J. A. (2008). Vis–NIR soluble solids prediction in intactoranges (Citrus sinensis L.) cv. Valencia Late by reflectance.Postharvest Biology and Technology, 47, 75–80.

Cayuela, J. S., & Weiland, C. (2010). Intact orange quality predictionwith two portable NIR spectrometers. Postharvest Biology andTechnology, 58, 113–120.

Cen, H., He, Y., & Huang, M. (2006). Measurements of soluble solidscontents and pH in orange juice using chemometrics and Vis–NIRS. Journal of Agricultural Food Chemistry, 54, 7437–7443.

Cen, H., Bao, Y., He, Y., & Sun, D.-W. (2007). Visible and nearinfrared spectroscopy for rapid detection of citric and tartaricacids in orange juice. Journal of Food Engineering, 82, 253–260.

Centner, V., Massart, D. L., de Noord, O. E., de Jong, S., Vandeginste, M.B., & Sterna, C. (1996). Elimination of uninformative variables formultivariate calibration. Analytical Chemistry, 68, 3851–3858.

Clark, C. J., McGlone, V. A., DeSilva, H. N., Manning, M. A.,Burdon, J., & Mowat, A. D. (2004). Prediction of storagedisorders of kiwifruit (Actanidia cinensis) based on visible-NIRspectral characteristics at harvest. Postharvest Biology andTechnology, 32, 147–158.

Cronje, P.J.R. (2005). Peteca spot of lemons. South African FruitJournal (Feb/March issue) 26–28

Cronje, P.J.R. (2009). Postharvest rind breakdown of ‘Nules Clem-entine’ mandarins (Citrus reticulate Blanco) fruit. Ph.D. thesis,Department of Horticultural Science, University of Stellenbosch,Stellenbosch, South Africa.

Cubero, S., Aleixos, N., Moltó, E., Gómez-Sanchis, J., & Blasco, J.(2010). Advances in machine vision applications for automaticinspection and quality evaluation of fruits and vegetables. Foodand Bioprocess Technology. doi:10.1007/s11947-010-0411-8.

De Jong, S. (1993). PLS fits closer than PCR. Journal of Chemometrics,7, 551–557.

Dull, G. G., & Birth, G. S. (1989). Nondestructive evaluation of fruitquality: use of near infrared spectrophotometry to measuresoluble solids in intact honeydew melons. HortScience, 24, 754.

Dull, G., Birth, G., & Leffler, R. (1989). Use of near infrared analysisfor the non-destructive measurement of dry matter in potatoes.American Potato Journal, 66, 215–225.

Dull, G. G., Birth, G. S., Smittle, D. A., & Leffler, R. G. (1989). Nearinfrared analysis of soluble solids of intact cantaloupe. Journal ofFood Science, 54, 393–395.

Fercher, A. F., Drexler, W., Hitzenberger, C. K., & Lasser, T. (2003).Optical coherence tomography—principles and applications.Reports on Progress in Physics, 66, 239–303.

Fraser, D. G., Künnemeyer, R., McGlone, V. A., & Jordan, R. B.(2001). Letter to the editor. Postharvest Biology and Technology,22, 191–195.

Fraser, D. G., Jordan, R. B., Künnemeyer, R., & McGlone, V. A.(2003). Light distribution inside mandarin fruit during internalquality assessment by NIR spectroscopy. Postharvest Biologyand Technology, 27, 185–196.

Fu, X., Ying, Y., Lu, H., & Xu, X. (2007). Comparison of diffusereflectance and transmission mode of visible-near infraredspectroscopy for detecting brown heart of pear. Journal of FoodEngineering, 83, 317–323.

Gaffney, J. J. (1973). Reflectance properties of citrus fruit. Trans-actions of the American Society of Agricultural Engineers, 16(2),310–314.

Geeola, F., Geeola, F., & Peiper, U. M. (1994). A spectrophotometricmethod for detecting surface bruises on ‘Golden Delicious’apples. Journal of Agricultural Engineering Research, 58, 47–51.

Golic, M., Walsh, K. B., & Lawson, P. (2003). Short-wavelength near-infrared spectra of sucrose, glucose, and fructose with respect tosugar concentration and temperature. Applied spectroscopy, 57,139–145.

Golic, M., & Walsh, K. B. (2006). Robustness of calibration medelsbased on near infrared spectroscopy for the in-line grading ofstonefruit for total soluble solids contents. Analytica ChimicaActa, 555, 286–291.

Gómez, A. H., He, Y., & Pereira, A. G. (2006). Non-destructivemeasurement of acidity, soluble solids and firmness of Satsumamandarin using vis–NIR spectroscopy techniques. Journal ofFood Engineering, 77, 313–319.

Gómez-Sanchis, J., Gómez-Chova, L., Aleixos, N., Camps-Valls, G.,Montesinos-Herrero, C., Moltó, E., et al. (2008). Hyperspectralsystem for early detection of rottenness caused by Penicilliumdigitatum in mandarins. Journal of Food Engineering, 89(1), 80–86.

Gowen, A. A., O’Donnell, C. P., Cullen, P. J., Downey, G., & Frias, J.M. (2007). Hyperspectral imaging-an emerging process analyticaltool for food quality and safety control. Trends in Food Science andTechnology, 18, 590–598.

Greensill, C. V., & Walsh, K. B. (2002). Calibration transfer betweenminiature photodiode array-based spectrometers in the nearinfrared assessment of mandarin soluble solids content. Journalof Near Infrared Spectroscopy, 10, 27–35.

Guthrie, J., & Walsh, K. (1997). Non-invasive assessment of pineappleand mango fruit quality using near infrared spectroscopy. AustralianJournal of Experimental Agriculture, 37, 253–263.

Guthrie, J. A., Wedding, B., & Walsh, K. B. (1998). Robustness ofNIR calibrations for soluble solids in intact melon and pineapple.Journal of Near Infrared Spectroscopy, 6, 259–265.

Guthrie, J. A., Walsh, K. B., Reid, D. J., & Liebenberg, C. J. (2005).Assessment of internal quality attributes of mandarin fruit. 1.NIR calibration model development. Australian Journal ofAgricultural Research, 56, 405–416.

Guthrie, J. A., Reid, D. J., & Walsh, K. B. (2005). Assessment ofinternal quality attributes of mandarin fruit. 2. NIR calibrationmodel robustness. Australian Journal of Agricultural Research,56, 417–426.

Guthrie, J. A., Liebenberg, C. J., & Walsh, K. B. (2006). NIR modeldevelopment and robustness in prediction of melon fruit total solublesolids. Australian Journal of Agricultural Research, 57, 1–8.

Hebden, J. C., Gibson, A., Yusof, R. M., Everdell, N., Hillman, E. M.C., Delpy, D. T., et al. (2002). Three-dimensional opticaltomography of the premature infant brain. Physics in Medicineand Biology, 47, 4155–4166.

Huang, D., Swanson, E. A., Lin, C. P., Schuman, J. S., Stinson, W. G.,Chang, W., et al. (1991). Optical coherence tomography. Science,254, 1178–1181.


http://dx.doi.org/10.1007/s11947-010-0411-8

Huang, H., Yu, H., Xu, H., & Ying, Y. (2008). Near infraredspectroscopy for on/in-line monitoring of quality in foods andbeverages: a review. Journal of Food Engineering, 87, 303–313.

Kader, A. A. (2002). Opportunities in using biotechnology to maintainpostharvest quality and safety of fresh produce. HortScience, 37,24–25.

Kawano, S., Fujiwara, T., & Iwamoto, M. (1993). Non-destructivedetermination of sugar content in ‘Satsuma’ mandarins usingNIRS transmittance. Journal of the Japanese Society forHorticultural Science, 62, 465–470.

Kemsely, E. K., Tapp, H. S., Binns, R., Mackin, R. O., & Peyton, A. J.(2008). Feasibility study of NIR diffuse optical tomography onagricultural produce. Postharvest Biology and Technology, 48,223–230.

Kim, J., Mowat, A., Poole, P., & Kasabov, N. (2000). Linear and non-linear pattern recognition models for classification of fruit fromvisible–near infrared spectra. Chemometrics and IntelligentLaboratory Systems, 51, 201–216.

Krivoshiev, G. P., Chalucova, R. P., & Moukarev, M. I. (2000). Apossibility for elimination of the interference from the peel innondestructive determination of the internal quality of fruit andvegetables by vis–NIR spectroscopy. Lebensm-Wiss University oftechnology, 33, 344–353.

Kutis, I. S., Sapozhnikova, V. V., Kuranov, R. V., & Kamenskii, V. A.(2005). Study of the morphological and functional state of higherplant tissues by optical coherence microscopy and opticalcoherence tomography. Russian Journal of Plant Physiology,52, 559–564.

Lammertyn, J., Peirs, J., De Baerdemaeker, J., & Nicolaï, B. M.(2000). Light penetration properties of NIR radiation in fruit withrespect to non-destructive quality assessment. Postharvest Biologyand Technology, 18, 121–132.

Lammertyn, J., Dressalaers, T., Van Hecke, P., Jancsók, P., Wevers,M., & Nicolaï, B. M. (2003). MRI and X-ray CT study of spatialdistribution of core breakdown in ‘Conference’ pears. MagneticResonance Imaging, 21(7), 805–815.

Lammertyn, J., Dresselaers, T., Van Hecke, P., Jancsók, P., Wevers,M., & Nicolaï, B. M. (2003). Analysis of the time course of corebreakdown in ‘Conference’ pears by means of MRI and X-rayCT. Postharvest Biology and Technology, 29, 19–28.

Leonardi, L., & Burns, D. H. (1999). Quantitative multiwavelengthconsistuent measurements using single-wavelength photo time-of-flight correction. Applied Spectroscopy, 53, 637–646.

Liu, Y., Sun, X., & Ouyang, A. (2010). Non-destructive measurementsof soluble solid content of navel orange fruit by visible-NIRspectrometric technique with PLS and PCA-BPNN. LWT-FoodScience and Technology, 43, 602–607.

Liu, Y., Sun, X., Zhang, H., & Aiguo, O. (2010). Nondestructivemeasurement of internal quality of Nanfeng mandarin fruit bycharge coupled device near infrared spectroscopy. Computersand Electronics in Agriculture, 71(S1), S10–S14.

López-García, F., Andreu-García, A., Blasco, J., Aleixos, N., &Valiente, J. M. (2010). Automatic detection of skin defects incitrus fruits using a multivariate image analysis approach.Computers and Electronics in Agriculture, 71, 189–197.

Louw, E. D., & Theron, K. I. (2010). Robust prediction models forquality parameters in Japanese plums (Prunus salicina L.) usingNIR spectroscopy. Postharvest Biology and Technology, 58, 176–184.

Lovász, T., Merész, P., & Salgó, A. (1994). Application of nearinfrared transmission spectroscopy for the determination of somequality parameters of apples. Journal of Near Infrared Spectroscopy,2, 213–221.

Lu, Q., Gan, X., Gu, M., & Luo, Q. (2004). Monte Carlo modelling ofoptical coherence tomography imaging through turbid media.Applied Optics, 43, 1628–1637.

Lu, H., Xu, H., Ying, Y., Fu, X., Yu, H., & Tian, H. (2006).Application Fourier transform near infrared spectrometer in rapidestimation of soluble solids content of intact citrus fruits. Journalof Zhejiang University Scence, 7, 794–799.

Magwaza, L.S., Opara, U.L., Nieuwoudt, H., & Cronje, P. (2011).Non-destructive quality assessment of ‘Valencia’ orange usingFT-NIR spectroscopy. In: Proceedings of the NIR 2011, 15thInternational Conference on Near-Infrared Spectroscopy, CapeTown, South Africa, 13–20 May 2011.

McClure, W. F. (2003). Review: 204 years of near infraredtechnology: 1800–2003. Journal of Near Infrared spectroscopy,11, 487–518.

McGlone, V. A., & Kawano, S. (1998). Firmness, dry-matter andsoluble solids assessment of postharvest kiwifruit by NIRspectroscopy. Postharvest Biology and Technology, 13, 131–141.

McGlone, V. A., Jordan, R. B., & Martinsen, P. J. (2002). Vis-NIRestimation at harvest of pre- and post-storage quality indices for‘Royal Gala’ apple. Postharvest Biology and Technology, 25,135–144.

McGlone, V. A., Fraser, D. G., Jordan, R. B., & Kunnemeyer, R. (2003).Internal quality assessment of mandarin fruit by vis–NIR spectroscopy.Journal of Near Infrared Spectroscopy, 11, 323–332.

Meglinski, I. V., Buranachai, C., & Terry, L. A. (2010). Plantphotonics: application of optical tomography to monitor defectsand rots in onion. Laser Physics Letters. doi:10.1002/lapl.200910141.

Mehl, P. M., Chen, Y.-R., Kim, M. S., & Chan, D. E. (2004).Development of hyperspectral imaging technique for the detectionof apple surface defects and contaminations. Journal of FoodEngineering, 61, 67–81.

Menesatti, P., Antonucci, F., Pallottino, F., Rocuzzo, G., Allegra, M.,Stagno, F., et al. (2010). Estimation of plant nutritional status byvis–NIR spectroscopic analysis on orange leaves [Citrus sinensis(L.) Osbeck cv Tarocco]. Biosystems Engineering, 105, 448–454.

Miller, B. K., & Delwiche, M. J. (1991). Spectral analysis of peachsurface defects. Transactions of the American Society forAgricultural Engineering, 34, 2509–2515.

Miller, W. M., & Zude, M. (2002). Non-destructive brix sensing ofFlorida grapefruit and honey tangerines. Proceedings of theFlorida state Horticultural society, 115, 56–60.

Miyamoto, K., & Kitano, Y. (1995). Non-destructive determination ofsugar content in Satsuma mandarin fruit by near infraredtransmittance spectroscopy. Journal of Near Infrared Spectroscopy,3, 227–237.

Miyamoto, K., Kawauchi, M., & Fukuda, T. (1998). Classification ofhigh acid fruits by PLS using the near infrared transmittancespectra of intact Satsuma mandarins. Journal of Near InfraredSpectroscopy, 6(1–4), 267–271.

Moon, D. G., & Mizutani, F. (2002). Relationship between fruit shapeacid content in different parts of citrus fruits. Journal of theJapanese Society for Horticultural Science, 71, 56–58.

Nicolaï, B. M., Lötze, E., Peirs, A., Scheerlinck, N., & Theron, K. I.(2006). Non-destructive measurement of bitter pit in apple fruitusing NIR hyperspectral imaging. Postharvest Biology andTechnology, 40, 1–6.

Nicolaï, B. M., Beullens, K., Bobelyn, E., Peirs, A., Saeys, W.,Theron, I. K., et al. (2007). Non-destructive measurement of fruitand vegetable quality by means of NIR spectroscopy: A review.Postharvest Biology and Technology, 46, 99–118.

Nicolaï, B. M., Theron, K. I., & Lammertyn, J. (2007). Kernel PLSregression on wavelet transformed NIR spectra for prediction ofsugar content of apple. Chemometrics and Intelligent LaboratorySystems, 85, 243–252.

Nicolaï, B. M., Verlinden, B. E., Desmet, M., Saevels, S., Saeys, W.,Theron, K., et al. (2008). Time-resolved and continuous waveNIR reflectance spectroscopy to predict soluble solids content


http://dx.doi.org/10.1002/lapl.200910141

http://dx.doi.org/10.1002/lapl.200910141

and firmness of pear. Postharvest Biology and Technology, 47,68–74.

Nicolaï, B. M., Bulens, I., De Baerdemaker, J., De Ketelaere, B., Hertog,M. L. A. T. M., Verboven, P., et al. (2009). Non-destructiveevaluation: detection of external and internal attributes frequentlyassociated with quality and damage. In D. Florkowiski (Ed.),Postharvest Handling: A Systems Approach (pp. 421–442).Amsterdam: Academic Press, Elsevier.

Nørgaard, L., Saudland, A., Wagner, J., Nielsen, J. P., Munck, L., &Engelsen, S. B. (2000). Interval partial least-square (iPLS): acomparative chemometric study with an example from nearinfrared spectroscopy. Applied Spectroscopy, 54, 413–419.

Norris, K. H., & Hart, J. R. (1965). Direct spectrophotometricdetermination of moisture content of grain and seeds. InHumidity and moisture, vol 4. Principles and methods ofmeasuring moisture in liquids and solids. New York: Reinhold.

Osborne, S. D., Jordan, R. B., & Kunnemeyer, R. (1997). Method ofwavelength selection for partial least squares. Analyst, 122,1531–1537.

Osborne, S. D., Kunnemeyer, R., & Jordan, R. B. (1999). A low-costsystem for the grading of kiwifruit. Journal of Near InfraredSpectroscopy, 7, 9–15.

Osborne, B. G. (2000). Near infrared spectroscopy in food analysis(pp. 1–14). Australia: BRI Australia Ltd.

Ou, A. S., Lin, S., Lin, T., Wu, S., & Tiarn, M. (1997). Studies on thedetermination of quality-related constituents in ‘Ponkan’ mandarinby near infrared spectroscopy. Journal of the Chinese AgriculturalChemical Society, 35, 462–474.

Pallav, P., Diamond, G. G., Hutchins, D. A., Green, R. J., & Gan, T. J.(2009). A near infrared (NIR) technique for imaging foodmaterials. Journal of Food Science, 74, 23–33.

Palmer, K. F., & Williams, D. (1974). Optical properties of water inthe near infrared. Journal of the Optical Society of America, 64,1107–1110.

Peirera, A. F. C., Pontes, M. J. C., Neto, F. F. G., Santos, S. R. B.,Galvaõ, R. K. H., & Araújo, M. C. U. (2008). NIR spectrometricdetermination of quality parameters in vegetable oils using iPLSand variable selection. Food Research International, 41, 341–348.

Peiris, K. H. S., Dull, G. G., Leffler, R. G., & Kays, S. J. (1998a). Near-infrared spectrometric method for nondestructive determination ofsoluble solids content of peaches. American society for HorticulturalScience, 123, 898–905.

Peiris, K. H. S., Dull, G. G., Leffler, R. G., & Kays, S. J. (1998b).Near-infrared (NIR) spectrometric technique for non-destructivedetermination of soluble solids content in processing tomatoes.American society for Horticultural Science, 123, 1089–1093.

Peiris, K. H. S., Dull, G. G., & Leffler, R. G. (1998c). Nondestructivedetection of selection drying, an internal disorder in tangerine.HortScience, 33, 310–312.

Peiris, K. H. S., Dull, G. G., Leffler, R. G., & Kays, S. J. (1999). Spatialvariability of soluble solids or dry-matter content within individualfruits, bulbs, or tubers: Implications for the development and use ofNIR spectrometric techniques. HortScience, 34, 114–118.

Peirs, A., Tirry, J., Verlinden, B., Darius, P., & Nicolaï, B. M. (2002).Effect of biological variability on the robustness of NIR-modelsfor soluble solids content of apples. Postharvest Biology andTechnology, 28, 269–280.

Peirs, A., Scheerlinck, N., & Nicolaï, B. M. (2003). Temperaturecompensation for near infrared reflectance measurement of applefruit soluble solids contents. Postharvest Biology and Technology,30, 233–248.

Qin, J., Burks, T. F., Kim, M. S., Chao, K., & Ritenour, M. A. (2008).Citrus canker detection using hyperspectral reflectance imagingand PCA-based image classification method. Sensory andInstrumental Food Quality, 2, 168–177.

Qin, J., Burks, T. F., Ritenour, M. A., & Bonn, W. G. (2009).Detection of citrus canker using hyperspectral reflectanceimaging with spectral information divergence. Journal of FoodEngineering, 93, 183–191.

Saiz-Abajo, M. J., Mevick, B.-H., Segtnan, V. H., & Naes, T. (2005).Ensemble methods and data augmentation by noise additionapplied to the analysis of spectroscopic data. Analytica ChimicaActa, 533, 147–159.

Sapozhnikova, V. V., Kamenskii, V. A., & Kuranov, R. V. (2003).Visualization of plant tissues by optical coherence tomography.Russian Journal of Plant Physiology, 50, 282–286.

Sapozhnikova, V. V., Kamensky, V. A., Kuranov, R. V., Kutis, I.,Snopova, L. B., & Myakov, A. V. (2004). In vivo visualization ofTradescantia leaf tissue and monitoring the physiological andmorphological states under different water supply conditionsusing optical coherence tomography. Planta, 219, 601–609.

Schaare, P. N., & Fraser, D. G. (2000). Comparison of reflectance andtransmission modes of visible-near infrared spectroscopy formeasuring internal properties of kiwifruit (Actinidia chinensis).Postharvest Biology and Technology, 20, 175–184.

Smith, L. G. (1984). Pineapple sepcifiuc gravity as an index of eatingquality. Tropical Agriculture (Trinidad), 61, 196–199.

Spreen, T.H. (2009). Projections of world production and consumptionof citrus to 2010. China/FAO citrus symposium. Food andAgricultural Organisation of the United nations. http://www.fao.org/docrep/003/x6732e/x6732e02.htm.

Steuer, B., Schulz, H., & Läger, E. (2001). Classification and analysisof citrus oils by NIR spectroscopy. Food Chemistry, 72, 113–117.

Sun, X., Zhang, H., & Liu, Y. (2009). Nondestructive assessment ofquality of ‘Nanfeng’ mandarin fruit by a portable near infraredspectroscopy. International Journal of Agricultural and BiologicalEngineering, 2, 65–71.

Tewari, J. C., Dixit, V., Chi, B.-K., &Malik, K. A. (2008). Determinationof origin and sugars of citrus fruit using genetic algorithm,correspondence analysis and partial least square combined withfiber optic NIR spectroscopy. Spectrochimica Acta Part A:Molecular and Biomolecular Spectroscopy, 71, 1119–1127.

Tomlins, P. H., & Wang, R. K. W. (2005). Theory, developments andapplications of optical coherence tomography. Journal ofPhysics. D: Applied Physics, 38, 2519–2535.

Tsuchikawa, S., Sakai, E., Inoue, K., & Miyamoto, K. (2003).Application of time-of-flight near-infrared spectroscopy to detectsugar and acid content in Satsuma mandarin. Journal of theAmerican Society for Horticultural Science, 128, 391–396.

Walsh, K. B., Guthrie, J. A., & Burney, J. W. (2000). Application ofcommercially available, low cost, miniaturised NIR spectrometers tothe assessment of the sugar content of intact fruit. AustralianJournal of Plant Physiology, 27, 1175–1186.

Walsh, K.B. (2005). Commercial adoption of technologies for fruitgrading, with emphasis on NIRS. Information and technology forsustainable fruit and vegetable production, FRUTIC 05, Montpellier,France, 12–16 September 2005

Wang, W., & Paliwal, J. (2007). Near-infrared spectroscopy andimaging in food quality and safety. Sensory and InstrumentalFood Chemistry, 1, 193–207.

Wetzel, D. L. (1983). Near infrared reflectance analysis: sleeperamong spectroscopic techniques. Analytical Chemistry, 55,1165–1176.

Williams, P. C., & Norris, K. H. (1987). Qualitative applications ofnear-infrared reflectance spectroscopy. In P. C. Williams & K. H.Norris (Eds.), Near-infrared technology in the agricultural andfood industries (pp. 241–246). St. Paul: American Association ofCereal Chemistry.

Williams, P., & Norris, K. H. (2001). Variable affecting near infraredspectroscopic analysis. In P. Williams & K. H. Norris (Eds.),Near infrared technology in the agriculture and food industries


http://www.fao.org/docrep/003/x6732e/x6732e02.htm

http://www.fao.org/docrep/003/x6732e/x6732e02.htm

(2nd ed., pp. 171–185). St Paul: The American Association ofCereal Chemists.

Wold, S., Sjostrom, M., & Eriksson, L. (2001). PLS-regression: abasic tool of chemometrics. Chemometrics and IntelligentLaboratory Systems, 58, 109–130.

Xia, J., Li, X., Li, P., Ma, Q., & Ding, X. (2007). Application ofwavelet transform in the prediction of ‘Navel’ orange vitamin Ccontent by near-infrared spectroscopy. Agricultural Sciences inChina, 6(9), 1067–1073.

Xing, J., Landahl, S., Lammertyn, J., Vrindts, E., & De Baerdemaeker,J. (2003). Effects of bruise type on discrimination of bruised andnonbruised ‘Golden Delicious’ apples by vis–NIR spectroscopy.Postharvest Biology and Technology, 30, 249–258.

Xing, J., Bravo, C., Jancsó, P. T., Ramon, H., & De Baerdemaeker, J.(2005). Detecting bruises on ‘Golden Delicious’ apples usinghyperspectral imaging with multiple wavebands. BiosystemsEngineering, 90, 27–36.

Zhao, X., Burks, T. F., Qin, J., & Ritenour, M. A. (2010). Effect of fruitharvest time on citrus canker detection using hyperspectral reflectanceimaging. Sensory and Instrumental Food Quality, 4, 126–135.

Zheng, Y., He, S., Yi, S., Zhou, Z., Mao, S., Zhao, X., et al. (2010).Predicting oleocellosis sensitivity in citrus using vis–NIRreflectance spectroscopy. Scientia Hoticulturae, 125, 401–405.

Zude, M., Pflanz, M., Kaprielian, C., & Aivazian, B. (2008). NIRS asa tool for precision horticulture in citrus industry. BiosystemEngineering, 99, 455–459.


NIR Spectroscopy Applications for Internal and External Quality Analysis of Citrus Fruit—A Review

Documents

Transcript of NIR Spectroscopy Applications for Internal and External Quality Analysis of Citrus Fruit—A Review