lable at ScienceDirect

Journal of Food Engineering 178 (2016) 151e158

Contents lists avai

Journal of Food Engineering

journal homepage: www.elsevier .com/locate/ j foodeng

Using sensor and spectral analysis to classify botanical origin anddetermine adulteration of raw honey

Zhilin Gan a, b, Yang Yang c, Jing Li d, Xin Wen a, b, Minghui Zhu a, b, Yundong Jiang a, b,Yuanying Ni a, b, *

a National Engineering Research Center for Fruits and Vegetables Processing, Beijing 100083, Chinab College of Food Science and Nutritional Engineering, China Agricultural University, Beijing 100083, Chinac Cloud Live Technology Group Co., Ltd., Beijing 100085, Chinad Institute of Electrical Engineering of the Chinese Academy of Sciences, Beijing 100190, China

a r t i c l e i n f o

Article history:Received 10 September 2015Received in revised form18 January 2016Accepted 19 January 2016Available online 21 January 2016

Keywords:HoneyElectronic noseElectronic tongueNIRMIR

* Corresponding author. College of Food ScienceChina Agricultural University, National Engineering RVegetables Processing, China.

E-mail address: niyy102@163.com (Y. Ni).

http://dx.doi.org/10.1016/j.jfoodeng.2016.01.0160260-8774/© 2016 Elsevier Ltd. All rights reserved.

a b s t r a c t

The feasibility of sensors (Electronic Nose, EN; Electronic Tongue, ET) and spectra (Near Infrared spec-trum, NIR; Mid Infrared spectrum, MIR) to evaluate raw honey samples (Vitex, Jujube and Acacia) wasexplored. Partial least squares discriminant analysis (PLSDA) model, support vector machine (SVM) al-gorithms model and Interval partial least squares (iPLS) model were used to classify the botanical origin.The results indicate that spectra and sensors could classify the botanical origin of honey rapidly andaccurately, since total accuracy for calibration and prediction sets was all almost 100% in EN and ETanalysis by SVM model and in NIR and MIR analysis by iPLS model. Then principal components analysis(PCA) and PLSDA model were used to determine the adulterants. Total accuracy for calibration andprediction sets was all above 96% in NIR, MIR and ET by PLSDA model. The results indicate that ET is moresuitable for detecting honey adulteration.

© 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Honey is a natural food product processed by honey bees byblending the sweetened sap collected from flowers with metabolicgastric enzymes (Anjos et al., 2015).

Natural honey is a very nutritious food product, which containswater (17% in general), saccharides (main constituents in honey,75% in general, mainly glucose and fructose), amino acids, minerals(Mg, Ca, K, Na, S, P, Fe, Mn, Co, Ni, etc.), vitamins, enzymes (inver-tase, catalase, amylase, etc.), phenols, organic acids, pigments,volatile oils and also over 100 varieties of aromatic substances (Dela Fuente et al., 2011; Mateo and Boschreig, 1997; Ouchemoukhet al., 2010; Arvanitoyannis et al., 2005; Baroni et al., 2006).

Honey needs to be rapidly evaluated and priced after collectionfrom a bee-house, based on botanical origin classification andadulterant determination. Honey has customarily been distin-guished as unifloral and multifloral depending on the botanical

and Nutritional Engineering,esearch Center for Fruits and

origin (Lenhardt et al., 2014). In China, over 20 kinds of unifloralhoney are commonly produced such as jujube, vitex, acacia, rape,and linden. In general, unifloral honey types have higher marketvalue due to their limited production and availability. Traditionally,botanical origin was determined by sensory analysis, pollenmorphology, characteristic aroma analysis and characteristic inte-rior components analysis. These methods have shown that botan-ical origin classification of honey is affected by color, aroma, contentof pollen, method of processing and storage, content of trace sub-stances and so on. However, these methods are expensive andtime-consuming. Although the results of sensory analysis are ac-curate, it is objective and needs a well trained certified taste panelwith a time investment. (Arvanitoyannis et al., 2005; Aparna andRajalakshmi, 1999; Lawal et al., 2009; Bogdanov et al., 2004;Bianchi et al., 2005; Iglesias et al., 2004; Martos et al., 2000;Gonzlez-Miret et al., 2005; Conti et al., 2007).

For unethical economic gain, honey adulteration is an obviousproblem in the market. Water, sucrose, inverted sugar, hydrox-ymethyl cellulose, dextrin and starch are adulterants which havebeen regularly identified by regular physicochemical analysis(Serrano et al., 2004). High performance liquid chromatography(HPLC), isotope mass spectrometry and capillary electrophoresis

Table 1Summaries of all the honey samples.

Z. Gan et al. / Journal of Food Engineering 178 (2016) 151e158152

have also been used to evaluate these. However, these methods arecomplicated, time and labor consuming (Morales et al., 2008; Tuet al., 2011; Luo et al., 2012). C-4 botanical glycosides like cornsyrup were added to honey subsequently to avoid detection. Eventhough corn syrup can be detected by stable carbon isotope ratioanalysis (SCIRA), it is too expensive and time consuming (Cotteet al., 2007). Moreover, SCIRA is not suitable for detection of C-3botanical glycoside such as rice syrup.

In recent years, rapid detection techniques such as spectraltechnology, sensor technology and chemical kits are widely used intesting of species, habitat, grades, freshness, nutrient quality, anddrug residues in agricultural products since they are time-saving,more convenient and accurate than traditional methods. Elec-tronic nose (EN) and electronic tongue (ET) are common methodsfor food analysis (Cozzolino et al., 2008; Wang et al., 2009;Ghasemi-Varnamkhasti et al., 2010; Vlasov et al., 2002; Lu et al.,2014; Pan et al., 2014; Haddi et al., 2013). A sample's whole infor-mation so called fingerprint data, can be determined from EN andET rather than qualification and quantification of some specificconstituents. Then a discriminant model will be established bycompiling fingerprint data and unknown samples could be deter-mined by the model. In previous studies, based on volatile com-ponents collected by solid phase micro-extraction, botanical originof honey has been detected byMass spectrum-electronic nose (MS-EN) (Ampuero et al., 2004). The bp-ANN model founded by ENconsisting of ten MOSFET and twelve MOS sensors could discrim-inate the botanical origin and geographical honey production area(Benedetti et al., 2004). aAstree-ET could classify botanical origin ofhoney as well, inwhich the accuracy of the artificial neural networkmodel was 100% (Major et al., 2011). Also PCA and ANN modelswere effective in classification of honey botanical origin by using ETwith metallic compound electrode (Escriche et al., 2012).

NIR and MIR are extensively performed in agricultural products,food and medicine analysis (Magwaza et al., 2011; Cen and He,2007; Balabin and Smirnov, 2011). Much research has revealedthat NIR could classify the botanical origin of honey by using PCAanalysis, canonical variate analysis, PLS discriminant analysis andlinear discriminant analysis (Davies et al., 2002; Ruoff et al., 2007).Some reports showed that NIR was also an effective method todiscriminate the adulterant of honey. Irish honey adulterated withbeet syrup and high fructose syrup has been correctly discrimi-nated and adulterant ratio could be predicted by using PLS analysis(Kelly et al., 2006a). Honey adulterated with different ratio offructose/glucose was also determined by PLSR, K-NN and SIMCAmodels (Downey et al., 2003). MIR could discriminate adulterant ofhoney as well. Honeywith glucose, fructose, sucrose and corn syrupadded was detected by MIR and the discriminant accuracy couldreach 90% by using LDA model (Irudayaraj et al., 2003). In anotherstudy, Irish artificial honey adulterated with fully inverted beetsyrup, high-fructose corn syrup, partially invert cane syrup,dextrose syrup and beet sucrose was evaluated by MIR and wellclassified by using SIMCA and PLSAD models (Kelly et al., 2006b).

Compared with previous studies, this work focused on evalu-ating the performance of sensors (EN and ET) and spectra (NIR andMIR) in botanical origin classification and adulterant determinationof raw honey. All honey samples used in this research were rawmaterials obtained directly from a bee-house and without anyprocessing. Different types of discriminant analysis models werebuilt. Sensors (ET and EN) and spectra (NIR and MIR) showeddifferent accuracy and performance in botanical origin classifica-tion and adulterants determination.

2. Materials and methods

2.1. Sample preparation

Three types of honey samples, vitex, jujube and acacia, wereprovided by Beijing Baihua Apiculture Technology DevelopmentCompany. Sample information is described in Table 1. Each sample,sourced from a different bee-house was filtered by 60 mesh screenin order to eliminate impurities. Samples were stored in the dark at4 �C until analysis.

In botanical origin classification, 105 samples were divided intotwo sets with calibration (79) and prediction (26) as shown inTable 2, respectively. Then 35 pure honey samples (including vitex,jujube and acaica) were chosen randomly to prepare adulterantsamples by adding two varieties of syrup (rice syrup and cornsyrup). Both syrups were purchased from Cargill Investments(China) Co., Ltd. The physicochemical properties of syrups andhoney samples were shown in Table 3. Adulterant samples wereprepared by mixing pure honey solution with syrup in differentconcentrations (5%, 10%, 20% and 40%). The samples, 259 in total,included 105 pure honey samples and 154 adulterated samples. Thedivision of sets was described as follow in Table 4.

2.2. Physical and chemical analysis of honey and syrup

The conductivity was measured by a Orion 5-Star BenchtopMeter (Thermo Fisher Scientific, Waltham, USA). The pH valueswere tested by a PB-10 pH meter (Sartorius Corp., Bohemia, NY,USA). Proline was evaluated by spectrophotometry. Saccharides,fructose, gluctose, cane sugar and maltose were determined byHPLC equipped with a YMC-Pack Polyamine II (250 mm � 4.6 mm,5 mm) column. All of the physiochemical properties above wereconducted by methods modified from previous studies (Bogdanovet al., 2004; Bogdanov, 2002).

2.3. Electronic nose

The EN system used for this study was a Heracles (Alpha Mos,Toulouse, France) which was equipped with a highly selective andsensitive specialty gas chromatograph. Volatile constituents ofhoney samples were collected by way of static headspace. Separa-tion columns used in this work were DB5 and DB1701, respectively.Retention time and peak area were detected by hydrogen flameionization detector (FIDs). Three grams of honey sample wasweighed into a septa-sealed screw-cap bottle and equilibrated for10 min at 65 �C (500 rpm). The parameters of EN were as follows:(1) Injector: temperature 180 �C, inject volume 3000 mL. (2) Tem-perature program: initial 40 �C, final 200 �C, heating rate 2.0 �C/s.(3) Trap temperature: 250 �C, purge time 15 s (4) Temperature ofFID 220 �C. (5) Acquisition time: 84 s (Yüzay and Selke, 2007). Alldeterminations reported were conducted in triplicate.

2.4. Electronic tongue

The sensor array of the a-Astree ET (Alpha Mos, Toulouse,France) was comprised of seven potentiometric chemical sensors

Variety Place of origin (different province of China) Quantity

Vitex Liao Ning, Hei Bei, Bei jing, Shan Xi 41Jujube Liao Ning, Hei Bei, Shan Dong, Shan Xi 27Acacia Shan Xi 37

Table 2Calibration and prediction sets of pure and adulterated honeys in botanical originclassification.

Type of samples Calibration set Prediction set Total

Vitex 31 10 41Jujube 20 7 27Acacia 28 9 41

Table 3Physicochemical properties of syrups and honey.

Conductivity(mS/cm)

pH Proline (mg/kg) Fructose (g/100 mL) Glucose (g/100 mL) Cane sugar (g/100 mL) Maltose (g/100 mL)

Corn syrup 20.41 4.45 46 32.66 37.32 nd ndRice syrup 15.53 4.54 41 42.60 33.32 0.21 ndHoney 265 4.43 293 39.55 28.91 1.49 1.44

Table 4Calibration and prediction sets of pure and adulterant honeys.

Type of samples Calibration set Prediction set Total

Pure honey samples 79 26 105Adulterated samples 115 39 154Adulterated 5% 22 8 30Adulterated 10% 35 12 47Adulterated 20% 35 11 46Adulterated 40% 23 8 31

Z. Gan et al. / Journal of Food Engineering 178 (2016) 151e158 153

(ZZ2808, JE5292, BB2011, CA5292, GA2808, HA2808, JB2808) andan Ag/AgCl standard electrode. The cross-selectivity of sensors wasfirst checked (Yang et al., 2013). The mass for each honey was 60 g,which was dissolved in an airtight glass jars with a volume of150 mL (concentration chamber), 80 mL sample each (Wei et al.,2009). The acquisition time of one evaluation was 120 s. Data wasrecorded every second. Then the mean value recorded between110 s and 120 s was determined. All determinations reported wereconducted in triplicate.

2.5. Near Infrared spectrum (NIR)

NIR spectra of the honey samples were obtained by using a FT-NIR system (Bruker Optics, Inc, Germany). Samples were scanned inthe region 10000e4000 cm�1 in transmittance mode by a FiberOptic Liquid Probe at room temperature. Each spectrum was ob-tained by averaging 64 scans. NIR spectral data were stored inabsorbance format with a resolution of 8 cm�1 (Chen et al., 2012).

2.6. Mid-infrared spectrum (MIR)

MIR spectra of honey samples were obtained by a FT-IR (PerkinElmer, USA) equipped with an Attenuated Total Reflection (ATR).Samples were scanned in the region 4000e650 cm�1 at roomtemperature. Each spectrum was obtained by averaging 4 scans.MIR spectral data were stored in absorbance format with a reso-lution of 4 cm�1 (Yusilawati et al., 2010).

2.7. Data analysis

All the pure honey and adulterant honey samples were detectedby EN, ET, NIR and MIR. Data analysis was performed by Matlab7.8.0. Before building chemometrics models, raw data were firstlypretreated (sensor data: standard normal variate (SNV) and auto-scale; spectra data: SNV, smoothing, auto-scale and derivatives).Partial least squares discriminant analysis (PLSDA) models were

built for botanical origin classification and adulterant determina-tion. Support vector machine discriminant analysis (SVMDA) andInterval Partial Least Squares (iPLS) were applied to optimize theresults of sensors (EN, ET) and spectra (NIR, MIR), respectively.Principal components analysis (PCA) was used to reduce the di-mensions of data. Outliers were eliminated before the foundationof every model.

3. Results

3.1. Physicochemical properties of honey and syrup

The mean values of the physicochemical parameters are shownin Table 3. Generally, the ratios and constituents of saccharides andpH were very similar in honey and syrup, while conductivity andproline were the more meaningful parameters to evaluate honeyand syrup due to the big difference. The difference in conductivity islikely due to natural honey's high mineral content. (Mateo andBoschreig, 1998).

3.2. Botanical origin classification

3.2.1. PLSDA model of EN, ET, NIR and MIRPLSDA model of EN was established based on discriminant

analysis results of seven principal components by using 20 sensors.PLSDA model of ET was found according to the results from

seven sensors' evaluation. Five principal components were thenchosen to optimize the results.

PLSDAmodel of NIR was built by discriminant analysis results often principle components.

PLSDA model of MIR was established by discriminant analysisresults of five principle components.

The PLSDA model results of these four methods were shown inTable 5. It was shown that the error rate was less than one in everymethod, both in calibration set and prediction set. In the MIR study,the accuracy of calibration and prediction reached 100% for allsamples. Compared with MIR, misjudgment and outliers were stillpresent in the NIR, EN and ET studies. The results indicated thatMIRperformed best using the PLSDA model.

3.2.2. SVMDA model of sensorsType C of SVMwas applied in this study and the kernel function

chosen was RBF (Radial basis function). Cost and gamma, the mainparameters of this function, were 100, 0.001 in EN and 100,0.0031623 in ET. Also SVs was 28 and 20 in EN and ET, respectively.Although the results shown in Fig. 1 confirmed that one Vitexsample and one Acacia sample which came from the calibration setmisjudged Acacia and Jujube in EN and ET, respectively, yet no er-rors occurred in prediction set. This means the SVMmodel can raisethe accuracy of botanical origin classification when compared withPLSDA model in EN and ET analysis. Traditionally, PCA, artificialneural network (ANN), linear discriminant analysis (LDA),discriminating factor analysis (DFA) were the best common modelsapplied to EN and ET data for identification of honey botanicalorigin (Huang et al., 2014; Ampuero et al., 2004; Tiwari et al., 2013;

Table 5Statistic results of discrimination using PLSDA based on NIR, MIR, EN and ET in botanical origin classification.

NIR MIR EN ET

Acacia Jujube Vitex Acacia Jujube Vitex Acacia Jujube Vitex Acacia Jujube Vitex

CalibrationCC 28 20 31 28 20 31 28 20 30 28 20 30WC 0 0 0 0 0 0 0 0 1 0 0 1CC% 100 100 100 100 100 100 100 100 96.77 100 100 96.77PredictionCC 8 6 9 9 7 10 9 7 9 8 6 10WC 1 1 1 0 0 0 0 0 1 1 1 0CC% 88.88 85.71 90 100 100 100 100 100 90 88.88 85.71 100

(C/W)C: (correct/wrong) classifications.

(a) (b)

Fig. 1. Plot for discriminating honey types by SVMDA based on EN and ET data. (a) EN (b) ET.

Z. Gan et al. / Journal of Food Engineering 178 (2016) 151e158154

Ulloa et al., 2013). Our work indicates that SVM is another potentialmodel as well.

3.2.3. iPLSDA model of spectraiPLSDA model was run to optimize the botanical origin classi-

fication results by using NIR and MIR. The spectra regions of NIRand MIR were divided into different sections. 6310-5847 cm�1 forNIR, 3397e3298 cm�1, 2893e2592 cm�1 and 1381e980 cm�1 forMIR were the optimized region. iPLSDA models of NIR and MIRwere found by discriminant analysis results of six principle com-ponents. The accuracy was 100% for both the calibration and pre-diction sets. Compared with a previous study (Chen et al., 2012)which used the BP-ANN model for NIR analysis, iPLSDA model inthis work more accurately distinguished the botanical origin ofAcacia, Vitex and Jujube. Fig. 2 is the Scatter plot for discriminatinghoney types by PLSDA based on NIR data ((a) Acacia (b) Jujue (c)Vitex). Only one error occurred in each type of honey. The resultsindicated that iPLS could optimize the spectra region whichreduced computational cost. Moreover, the discrimination amongthe samples was improved due to the more obvious distribution ofdifferent samples. In the optimized region of NIR, OeH secondarymultiple frequency, CeH secondary multiple frequency and com-bination band were the predominant characteristics. CeH, OeH,H2O, CeO and CeC stretching bands as well as OeCeH, CeCeH andCeOeH were the prominent information in MIR.

3.3. Adulteration determination

3.3.1. PCA analysisThe PCA scatter plot of EN is shown in Fig. 3 (a). Adulterated

samples and pure honey samples could not be well discriminatedby EN due to the crossover although they were separated to somedegree. Also it was inferred that obvious difference of volatilecomponents among part of the samples existed because of a fewdiscrete points. There are two reasons for this difference: (1)constitution of volatile substance is different in three kinds of purehoney, even in different samples of the same botanical origin. Everysample was derived from different bee-house, so it could beinfluenced by the natural environment. (2) The fragrance charac-teristic has been changed after adding the syrup. So it is difficult forEN to classify adulterant samples since the volatile components arecomplicated.

ET-PCA which is described in Fig. 3(b) shows that pure honeycould be divided into three groups according to different botanicalorigin. The corresponding adulterants were discriminated as threegroups as well. Additionally, pure honey and adulterant could beclearly distinguished by ET.

As shown in Fig. 3(c), pure honey and adulterant were uniformlydistributed leading to NIR-PCA's inability to classify adulteration ofhoney.

Both division and crossover existed in Fig. 3(d). This means purehoney and adulterant could not be accurately discriminated byMIRin PCA analysis.

It was concluded that honey samples of this study could bebetter discriminated by ET, compared using EN, NIR and MIR in PCA

Fig. 2. Scatter plot for discriminating honey types by PLSDA based on NIR data. (a) Acacia; (b) Jujube; (c) Vitex.

Z. Gan et al. / Journal of Food Engineering 178 (2016) 151e158 155

analysis, nomatter whether among adulterant samples, or betweenpure honey and adulterant. Saccharide intermolecular bonds werethe predominant bond type present in the spectra. Based on theresult of 3.1, the ratio and constituent of saccharides and pH werevery similar in honey and syrup. So adulterant could not bedetermined by PCA analysis in NIR and MIR. More pretreatment ofdata and foundation of discriminant models are needed. Combinedwith 3.1, difference of conductivity between pure honey and syrupis mainly because more minerals exist in pure honey. ET is moresensitive to minerals. Furthermore, there is gustatory differencebetween pure honey and adulterants which can be exactly classi-fied by the sensors of ET. In brief, ET is more suitable for detectinghoney adulteration.

3.3.2. PLSDA modelsFewer than 105 and 154 pure honey and adulterant samples,

respectively, were usable due to outliers.The results from the PLSDA model for EN are shown in Table 6.

Six principle components were collected in EN analysis and outlierswere eliminated. Misjudgment still existed in calibration and pre-diction. Thus, EN was probably not suitable to determine theadulteration of honey based on the PLSDA model. The reason couldbe as follows: (1) Three different kinds of honey collected in this

study were substantially different in terms of volatile components.(2) Even samples of same origin which were collected fromdifferent area were obviously different according to the EN foot-prints. (3) The “smell” affected by freshness of samples wasdifferent due to the collecting and processing time.

ET-PLSDA model was established by five principle componentsafter eliminating three abnormal adulterant samples. The totaldiscriminant accuracy of calibration and prediction were 98.43%and 100%, respectively, as shown in Table 6. That means adulterantcan be accurately determined by combining the ET-PLSDA modelwith the results from ET-PCA, which can be explained by syrup'slack of trace substances such as acids, phenols, enzymes, mineralsand amino acids neither such physicochemical properties like pH,conductivity which all exist in honey and are easily detected by ET.However, a previous study (Zakaria et al., 2011), demonstrated thatsingle modality assessment of EN and ET was ineffective todiscriminate the adulterated honey by PCA and LDA.

Table 7 shows PLSDA results for discriminating honey adulter-ation using NIR and MIR by pretreatment of SNVþ2nd deriva-tiveþ15point smooth þ auto-scale. Except for one misjudgmentthat appeared in calibration set of MIR, the accuracy of other sets allreached 100% as shown in Table 7. SNVþ2nd derivativeþ15pointsmooth þ auto-scale performed best and was selected to build the

Fig. 3. PCA score plot of pure and adulterant honeys based on EN, ET, NIR and MIR data. (a) EN (b) ET (c) NIR (d) MIR.

Table 6PLSDA results for discriminating honey adulteration using EN and ET data.

Type of samples EN ET

Calibration Prediction Calibration Prediction

Recognition ratio % Misjudged/Total Recognition ratio % Misjudged/Total Recognition ratio % Misjudged/Total Recognition ratio % Misjudged/Total

Pure honey 95.95 3/74 98.72 1/25 97.47 2/79 100 0/26Adulterated total 93.52 7/108 80.56 7/36 99.11 1/112 100 0/39Adulterated 5% 95.24 1/21 71.42 2/7 95.45 1/22 100 0/8Adulterated 10% 93.94 2/33 81.82 2/11 100 0/35 100 0/12Adulterated 20% 87.88 4/33 80 2/10 100 0/35 100 0/11Adulterated 40% 100 0/21 87.5 1/8 100 1/23 100 0/8

Table 7PLSDA results for discriminating honey adulteration using NIR and MIR data.

Type of samples NIR MIR

Calibration Prediction Calibration Prediction

Recognition ratio % Misjudged/Total Recognition ratio % Misjudged/Total Recognition ratio % Misjudged/Total Recognition ratio % Misjudged/Total

Pure honey 100 0/79 100 0/26 100 1/79 96.15 1/26Adulterated total 100 0/115 100 0/39 99.13 1/115 100 0/39Adulterated 5% 100 0/22 100 0/8 94.45 1/22 100 0/8Adulterated 10% 100 0/35 100 0/12 100 0/35 100 0/12Adulterated 20% 100 0/35 100 0/11 100 0/35 100 0/11Adulterated 40% 100 0/23 100 0/8 100 0/23 100 0/8

Z. Gan et al. / Journal of Food Engineering 178 (2016) 151e158156

Z. Gan et al. / Journal of Food Engineering 178 (2016) 151e158 157

models compared to SNV þ auto-scale and SNVþ1st deriva-tiveþ15point smooth þ auto-scale. It is probably because 2nd de-rivative can amplify the difference between adulterated honey andpure honey. Thus, PLSDA model is suitable for NIR and MIR. In aprevious study (Zhu et al., 2010), NIR combined with chemometricsmethods has been used to detect adulteration of honey samples.Least square support vector machine (LS-SVM) model was estab-lished and the total accuracy was 95.1%. Our model has more po-tential due to its higher accuracy.

4. Conclusions

Botanical origin of honey could be rapidly determined by EN, ET,NIR and MIR. Compared with PLSDA, iPLS and SVM could improvethe accuracy of spectra and sensors. It was found that spectra arebetter than sensors due to higher accuracy. Additionally, samplesneed pre-treatment in sensors analysis. For instance, aroma ofhoney should be equilibrated in EN and samples need to be dilutedin ET. Although sensor technology is better than traditionalmethods, spectra will probably be more frequently applied inbotanical origin classification for honey in real time and on-site.

The performed tests show that ET, NIR and MIR could be used inadulterant determination of honey. EN is more sensitive to aromasubstances and suitable for testing freshness of honey. Adulteratedhoney could not be determined by PCA in NIR and MIR analysis.After pretreatment of second derivative, NIR and MIR could accu-rately determine the adulteration in honey. ET performed best inPCA analysis. This method is easier and more rapid to test andanalyze. Data processing is simpler when using ET as well. Addi-tionally, ET is more sensitive to substances like amino acids, min-erals, phenols, monosaccharides and disaccharides which exist inhoney as shownwith NIR andMIR. So it can be inferred that ETmaylead to more extensive application for determination of honeyadulterants. Further research is needed because few studies havepublished on this topic.

Acknowledgement

We thank the Beijing Baihua Apiculture Technology Develop-ment Company for providing the raw honey. Also, we are grateful toProf. Han, Dr. Zhou, and Dr. Lv, who all work in College of Engi-neering, China Agricultural University, and Bruker Optics, Inc,Germany, for guiding us in the use of NIR andMIR.We acknowledgefunding of Natural Science Foundation of China (NO. 31171675) andSpecial Fund for Agro-scientific Research in the Public Interest (No.201503142-01) from the Ministry of Agriculture of the People'sRepublic of China.

References

Ampuero, S., Bogdanov, S., Bosset, J.O., 2004. Classification of unifloral honeys withan MS-based electronic nose using different sampling modes: SHS, SPME andINDEX. Eur. Food Res. Technol. 218, 198e207.

Anjos, O., Iglesias, C., Peres, F., Martinez, J., Garcia, A., Taboada, J., 2015. Neuralnetworks applied to discriminate botanical origin of honeys. Food Chem. 175,128e136.

Aparna, A.R., Rajalakshmi, D., 1999. Honeydits characteristics, sensory aspects, andapplications. Food Rev. Int. 15, 455e471.

Arvanitoyannis, I.S., Chalhoub, C., Gotsiou, P., Lydakis-Simantiris, N., Kefalas, P.,2005. Novel quality control methods in conjunction with chemometrics(Multivariate analysis) for detecting honey authenticity. Crit. Rev. Food Sci. Nutr.45, 193e203.

Balabin, R.M., Smirnov, S.V., 2011. Melamine detection by mid- and near-infrared(MIR/NIR) spectroscopy: a 371 quick and sensitive method for dairy productsanalysis including liquid milk, infant formula, and milk powder. Talanta 85,562e568.

Baroni, M.V., Nores, M.L., Díaz, M.D.P., Chiabrando, G.A., Fassano, J.P., Costa, C.,Wunderlin, D.A., 2006. Determination of volatile organic compound patternscharacteristic of five unifloral honey by solid-phase microextraction-gas

chromatography-mass spectrometry coupled to chemometrics. J. Agric. FoodChem. 54, 7235e7241.

Benedetti, S., Mannino, S., Sabatini, A.G., Marcazzan, G.L., 2004. Electronic nose andneural network use for the classification of honey. Apidologie 35, 397e402.

Bianchi, F., Careri, M., Musci, M., 2005. Volatile norisoprenoids as markers ofbotanical origin of Sardinian strawberry-tree (Arbutus unedo L.) honey: char-acterisation of aroma compounds by dynamic headspace extraction and gaschromatographyemass spectrometry. Food Chem. 89, 527e532.

Bogdanov, S., 2002. Harmonised methods of the international honey commission.Swiss Bee research Centre, FAM, Liebefeld, CH-3003 Bern, Switzerland.

Bogdanov, S., Ruoff, K., Persano Oddo, L., 2004. Physico-chemical methods for thecharacterisation of unifloral honeys: a review. Apidologie 35, S4eS17.

Cen, H., He, Y., 2007. Theory and application of near infrared reflectance spectros-copy in determination of food quality. Trends Food Sci. Technol. 18, 72e83.

Chen, L., Wang, J., Ye, Z., Zhao, J., Xue, X., Vander Heyden, Y., Sun, Q., 2012. Classi-fication of Chinese honeys according to their floral origin by near infraredspectroscopy. Food Chem. 135, 338e342.

Conti, M.E., Stripeikis, J., Campanella, L., Cucina, D., Tudino, M.B., 2007. Character-ization of Italian honeys (Marche Region) on the basis of their mineral contentand some typical quality parameters. Chem. Cent. J. 1, 14.

Cotte, J.F., Casabianca, H., Lh�eritier, J., Perrucchietti, C., Sanglar, C., Waton, H., Gre-nier-Loustalot, M.F., 2007. Study and validity of 13C stable carbon isotopic ratioanalysis by mass spectrometry and 2H site-specific natural isotopic fraction-ation by nuclear magnetic resonance isotopic measurements to characterizeand control the authenticity of honey. Anal. Chim. Acta 582, 125e136.

Cozzolino, D., Smyth, H.E., Cynkar, W., Janik, L., Dambergs, R.G., Gishen, M., 2008.Use of direct headspace-mass spectrometry coupled with chemometrics topredict aroma properties in Australian riesling wine. Anal. Chim. Acta 621, 2e7.

Davies, A.M.C., Radovic, B., Fearn, T., Anklam, E., 2002. A preliminary study on thecharacterisation of honey by near infrared spectroscopy. J. Near InfraredSpectrosc. 10 (2), 121e135.

De la Fuente, E., Ruiz-Matute, A.I., Valencia-Barrera, R.M., Sanz, J., Martínez Castro, I.,2011. Carbohydrate composition of Spanish unifloral honeys. Food Chem. 129,1483e1489.

Downey, G., Fouratier, V., Kelly, J.D., 2003. Detection of honey adulteration byaddition of fructose and glucose using near infrared transflectance spectros-copy. J. Near Infrared Spectrosc. 11, 447e456.

Escriche, I., Kadar, M., Domenech, E., Gil-S�anchez, L., 2012. A potentiometric elec-tronic tongue for the discrimination of honey according to the botanical origin.Comparison with traditional methodologies: physicochemical parameters andvolatile profile. J. Food Eng. 109, 449e456.

Ghasemi-Varnamkhasti, M., Mohtasebi, S.S., Siadat, M., 2010. Biomimetic-basedodor and taste sensing systems to food quality and safety characterization: anoverview on basic principles and recent achievements. J. Food Eng. 100,377e387.

Gonzlez-Miret, M.L., Terrab, A., Hernanz, D., Fernandez-Recamales, M.A.,Heredia, F.J., 2005. Multivariate correlation between color and mineralcomposition of honeys and by their botanical origin. J. Agric. Food Chem. 53 (7),2574e2580.

Haddi, Z., Alami, H., El Bari, N., Tounsi, M., Barhoumi, H., Maaref, A., Jaffrezic-Renault, N., Bouchikhi, B., 2013. Electronic nose and tongue combination forimproved classification of Moroccan virgin olive oil profiles. Food Res. Int. 54,1488e1498.

Huang, L., Liu, H., Zhang, B., Wu, D., 2014. Application of electronic nose withmultivariate analysis and sensor selection for botanical origin identification andquality determination of honey. Food Bioprocess Technol. 8, 359e370.

Iglesias, M.T., De Lorenzo, C., Polo, M.D.C., Martín-�Alvarez, P.J., Pueyo, E., 2004.Usefulness of amino acid composition to discriminate between Honeydew andFloral honeys. Application to honeys from a small Geographic area. J. Agric. FoodChem. 52, 84e89.

Irudayaraj, J., Xu, F., Tewari, J., 2003. Rapid determination of invert cane sugaradulteration in honey using FTIR spectroscopy and multivariate analysis. FoodEng. Phys. Prop. 68, 2040e2045.

Kelly, J.D., Petisco, C., Downey, G., 2006a. Potential of near infrared transflectancespectroscopy to detect adulteration of Irish honey by beet invert syrup and highfructose corn syrup. J. Near Infrared Spectrosc. 14 (2), 139e146.

Kelly, J.D., Petisco, C., Downey, G., 2006b. Application of Fourier transform mid-infrared spectroscopy to the discrimination between Irish artisanal honey andsuch honey adulterated with various sugar syrups. J. Agric. Food Chem. 54,6166e6171.

Lawal, R.A., Lawal, A.K., Adekalu, J.B., 2009. Physico-chemical studies on adultera-tion of honey in Nigeria. Pak. J. Biol. Sci. 12 (15), 1080e1084.

Lenhardt, L., Zekovi�c, I., Drami�canin, T., Te�si�c, �Z., Milojkovi�c-Opsenica, D.,Drami�canin, M.D., 2014. Authentication of the botanical origin of unifloralhoney by infrared spectroscopy coupled with support vector machine algo-rithm. Phys. Scr. T162 (014042), 1e4.

Lu, L., Deng, S., Zhu, Z., Tian, S., 2014. Classification of Rice by combining electronictongue and nose. Food Anal. Methods 1893e1902.

Luo, D., Luo, H., Xian, Y., Wang, B., Wu, W., Chen, Y., Guo, X., Wu, Y., 2012. Identi-fication the authenticity of no protein honey by IRMS. Mod. Food Sci. Technol.28, 862e866.

Magwaza, L.S., Opara, U.L., Nieuwoudt, H., Cronje, P.J.R., Saeys, W., Nicolaï, B., 2011.NIR spectroscopy applications for internal and external quality analysis of CitrusFruitdA review. Food Bioprocess Technol. 5, 425e444.

Major, N., Markovic, K., Krpan, M., Saric, G., Hruskar, M., Vahcic, N., 2011. Rapid

Z. Gan et al. / Journal of Food Engineering 178 (2016) 151e158158

honey characterization and botanical classification by an electronic tongue.Talanta 85, 569e574.

Martos, I., Ferreres, F., Tom�as-Barber�an, F.A., 2000. Identification of flavonoidmarkers for the botanical origin of Eucalyptus honey. J. Agric. Food Chem. 48,1498e1502.

Mateo, R., Boschreig, F., 1997. Sugar profiles of Spanish unifloral honeys. Food Chem.60 (1), 33e41.

Mateo, R., Boschreig, F., 1998. Classification of Spanish unifloral honeys bydiscriminant analysis of electrical conductivity, color, water content, sugars, andpH. J. Agric. Food Chem. 46 (2), 393e400.

Morales, V., Corzo, N., Sanz, M.L., 2008. HPAEC-PAD oligosaccharide analysis todetect adulterations of honey with sugar syrups. Food Chem. 107, 922e928.

Ouchemoukh, S., Schweitzer, P., Bachir Bey, M., Djoudad-Kadji, H., Louaileche, H.,2010. HPLC sugar profiles of Algerian honeys. Food Chem. 121, 561e568.

Pan, L., Zhang, W., Zhu, N., Mao, S., Tu, K., 2014. Early detection and classification ofpathogenic fungal disease in post-harvest strawberry fruit by electronic noseand gas chromatography-mass spectrometry. Food Res. Int. 62, 162e168.

Ruoff, K., Luginbühl, W., Bogdanov, S., Bosset, J.-O., Estermann, B., Ziolko, T.,Kheradmandan, S., Amad, R., 2007. Quantitative determination of physical andchemical measurands in honey by near-infrared spectrometry. Eur. Food Res.Technol. 225, 415e423.

Serrano, S., Villarejo, M., Espejo, R., Jodral, M., 2004. Chemical and physical pa-rameters of Andalusian honey: classification of Citrus and Eucalyptus honeys bydiscriminant analysis. Food Chem. 87, 619e625.

Tiwari, K., Tudu, B., Bandyopadhyay, R., Chatterjee, A., 2013. Identification ofmonofloral honey using voltammetric electronic tongue. J. Food Eng. 117,205e210.

Tu, Z., Zhu, D., Ji, B., Chen, H., Qing, Z., 2011. Adulteration detection of honey based

on near-infrared spectroscopy. Trans. Chin. Soc. Agric. Eng. 27, 382e388.Ulloa, P.A., Guerra, R., Cavaco, A.M., Rosa da Costa, A.M., Figueira, A.C., Brigas, A.F.,

2013. Determination of the botanical origin of honey by sensor fusion ofimpedance e-tongue and optical spectroscopy. Comput. Electron. Agric. 94,1e11.

Vlasov, Y., Legin, A., Rudnitskaya, A., 2002. Electronic tongues and their analyticalapplication. Anal. Bioanal. Chem. 373, 136e146.

Wang, Y., Wang, J., Zhou, B., Lu, Q., 2009. Monitoring storage time and qualityattribute of egg based on electronic nose. Anal. Chim. Acta 650, 183e188.

Wei, Z., Wang, J., Liao, W., 2009. Technique potential for classification of honey byelectronic tongue. J. Food Eng. 94, 260e266.

Yang, Y., Chen, Q., Shen, C., Zhang, S., Gan, Z., Hu, R., Zhao, J., Ni, Y., 2013. Evaluationof monosodium glutamate, disodium inosinate and guanylate umami taste byan electronic tongue. J. Food Eng. 116, 627e632.

Yusilawati, A.N., Maizirwan, M., Hamzah, M.S., Ng, K.H., Wong, C.S., Islamic, I.,Lumpur, K., 2010. Surface modification of polystyrene beads by Ultraviolet/Ozone treatment and its effect on gelatin coating department of biotechnologyengineering. Am. J. Appl. Sci. 7, 724e731.

Yüzay, B.I.E., Selke, S., 2007. Development of Electronic Nose Method for Evaluationof Residual Solvents in Low-density Polyethylene Films. Pack. Technol. Sci. 20,99e112.

Zakaria, A., Shakaff, A.Y., Masnan, M.J., Ahmad, M.N., Adom, A.H., Jaafar, M.N.,Ghani, S.A., Abdullah, A.H., Aziz, A.H., Kamarudin, L.M., Subari, N., Fikri, N.A.,2011. A biomimetic sensor for the classification of honeys of different floralorigin and the detection of adulteration. Sensors 11, 7799e7822.

Zhu, X., Li, S., Shan, Y., Zhang, Z., Li, G., Su, D., Liu, F., 2010. Detection of adulterantssuch as sweeteners materials in honey using near-infrared spectroscopy andchemometrics. J. Food Eng. 101, 92e97.