Ee

Aa

b

a

AA

KHPDO

1

uiptqcisvvbpaci

dpio[

0h

Sensors and Actuators B 187 (2013) 234– 240

Contents lists available at SciVerse ScienceDirect

Sensors and Actuators B: Chemical

j o ur n al hom epa ge: www.elsev ier .com/ locate /snb

valuation of organoleptic and texture properties of dried apples by hybridlectronic tongue

nna Kutyła-Olesiuka, Małgorzata Nowackab, Małgorzata Wesołya, Patrycja Cioseka,∗

Warsaw University of Technology, Faculty of Chemistry, Department of Microbioanalytics, Noakowskiego 3, 00-664 Warsaw, PolandWarsaw University of Life Sciences, Faculty of Food Sciences, Department of Food Engineering and Process Management, Nowoursynowska 159c, 02-776 Warsaw, Poland

r t i c l e i n f o

rticle history:vailable online 6 November 2012

eywords:ybrid electronic tongue

a b s t r a c t

The aim of this paper is to use hybrid electronic tongue in the qualitative and quantitative analysis ofextracts obtained from raw and dried apples prepared by different food processing techniques (dryingtechniques). The system is based on potentiometric and voltammetric sensors, moreover data recordedby spectrophotometry, amperometry and conductometry techniques were applied to enhance the classi-

LS-DAried applesrganoleptic analysis

fication ability of the device. The combination of the data from various measurement techniques (hybridelectronic tongue) leads to improved differentiation of the dried apple extracts samples comparing toseparate techniques. Appropriate chemometric techniques were used to the recognition and classifica-tion of samples, whereas the efficiency of the qualitative and quantitative analysis was evaluated onthe basis of Root Mean Squared Error (RMSE), determination coefficient (R2), slope (a), and intercept (b)

parameters.

. Introduction

In the era of expanding the production of ready meals and prod-cts for direct consumption the drying of fruits and vegetables is

mportant. Although drying is one of the oldest methods of foodreservation, the process is widely used and it is oriented on con-inuous development [1–4]. Fruits and vegetables contain a largeuantity of initial moisture content and are therefore highly sus-eptible to rapid quality degradation, even to the extent of spoilage,f not kept in thermally controlled storage facilities. Moreovereasonality of plant materials often leads to a surplus of fruit andegetable, and drying makes it possible to easily and quickly con-ert these perishable products into more stabilized ones that cane kept under a minimal controlled environment for an extendederiod of time [5]. On the other hand, drying changes the physicalnd biochemical form of the fruit leading to shrinkage and change ofolor, texture, taste, and so on [6,7]. These properties are importantn selecting a product by the consumer.

Color is an important indicator of food quality. The color of theried product has a strong influence on the acceptability of theroduct by the consumer, because the first assessment of the food

s made mainly based on visual impression [8–10]. The methods

f drying and process parameters have a significant effect on it8]. During the drying of plant products color changes occur as a

∗ Corresponding author. Tel.: +48 22 2347873; fax: +48 22 2347873.E-mail address: pciosek@ch.pw.edu.pl (P. Ciosek).

925-4005/$ – see front matter © 2012 Elsevier B.V. All rights reserved.ttp://dx.doi.org/10.1016/j.snb.2012.10.133

© 2012 Elsevier B.V. All rights reserved.

result of the browning reaction, both enzymatic and non-enzymatic[11,12].

Unfortunately, convective drying causes adverse change ofcolor. With the increase of temperature of drying air and dryingtime, the color becomes darker [9,13]. Similarly, infrared dryingoften causes darkening of the tissue as a result of the browningreaction [14]. However, using the microwave power for dryinggives the product with much better color than convective drying.Microwaves cause small color changes [15,16]. Moreover, dryingeffect on tissue texture changes such as firmness and hardnessof the material [17]. Loss of the fragility of foods is associatedwith increased water content, which is a plasticizer [18]. With theincrease of water activity decreases mechanical strength of theproduct. Texture also significantly affects the acceptability of theproduct by the consumer [11].

Both the color and texture of dried tissue (which are measuredwith physical methods), affect the sensory evaluation made byconsumers. However, the correlation is not straightforward andobvious, therefore taste and quality must be judged with humanpanels, which is a time-consuming, subjective, and costly method.That is the main reason for conducting research focused on estima-tion of organoleptic properties of food with objective methods. Oneof the most promising tools used in this area are sensor arrays cou-pled with pattern recognition blocks, i.e. electronic tongues [19].

An electronic tongue was developed as a device inspired by

the natural sense of taste [20]. Of course, nowadays systemsperformance is a very far analogy of natural counterparts, despiteof that the electronic tongues can be applied successfully inspecific applications. The use of such systems enables to perform

and A

tTtbp

aowymevttb

twssPfbwv

psatfo

2

2

oFo0rep

pflwc

dmrvww

edpfl1i

A. Kutyła-Olesiuk et al. / Sensors

he recognition and classification of sample images [19,21,22].he number of papers describing the application of electronicongue for the measurement of environmental samples [23],iotechnological samples [24], food [25], body fluids [26] andharmaceuticals [27] is steadily growing.

Commonly, potentiometric and voltammetric sensors arepplied to build the sensor array of an electronic tongue, but alsoptical, amperometric, and conductometric sensors are involved asell [21]. Pattern recognition system enabling appropriate anal-

sis of sensor array responses comprises various chemometricethods including neural networks. In last years, so-called hybrid

lectronic tongues have become more popular. They are based onarious sensing schemes, which allows for obtaining more informa-ion about the samples than separately used techniques, which, inurn, enables more accurate characteristics and clearer distinctionetween them [28].

The literature reports the application of electronic tongues forhe analysis of many foodstuffs: non-alcoholic beverages [29],ater [30], juices [31], etc. With the use of electronic tongue and

ensory evaluation apple juice samples and modified juices weretudied for apple taste, sweet taste and sour taste estimation [32].otentiometric electronic tongue was applied to distinguish dif-erent fruit juices [33]. Following example is an electronic tongueased on potentiometric sensors, HPLC and FTIR spectroscopy,hich was used to the qualitative and quantitative analysis of apple

arieties [34].In this work we propose a hybrid electronic tongue based on

otentiometry, voltammetry, amperometry, conductometry andpectroscopy for the evaluation of organoleptic properties of driedpples. Moreover, the correlation of sensor array responses withaste, color and texture properties is studied, which could be use-ul for future quality control, estimation of nutritional value, andrganoleptic evaluation purposes.

. Experimental

.1. Preparation of dried apples

The experimental material comprised apples (var. Idared). Itriginated from the Department of Fruit Culture Experimentalields of SGGW (Warsaw University of Life Sciences) and wasbtained in December. Apples were peeled and cut into slices.005 ± 0.001 m thick and diameter of 0.03 m. The cut mate-ial was immersed in the 0.1% citric acid solution to preventnzymatic browning reactions and it was blotted with filteraper.

Convective drying was performed in a prototype dryer. Sam-les were placed on a perforated tray positioned co-current to airow. Air velocity was 2 m/s and temperature was 70 ◦C. The dryeras loaded with 1.92 kg/m2 of apple tissue. The drying process was

arried on until constant mass was achieved.Microwave-convective drying was performer in a laboratory

ryer fabricated by Promise Tech Inc. (Wroclaw, Poland) withicrowave incident power of 300 W. Samples were placed on a

otating tray on shelves positioned perpendicular to air flow. Airelocity was 3.5 m/s and temperature 40 ◦C. The dryer was loadedith 2.4 kg/m2 of apple tissue. Drying was run until about 10%ater in the dried material was reached.

Infrared drying was performed in a dryer equipped with infraredlectric bulbs. The infrared emitters were located 0.2 m from theried product. The total power of emitters was 7.875 kW/m2. Sam-

les were placed on a perforated tray positioned co-current to airow. The flow of ambient air over the heated surface was set at.2 m/s. The dryer was loaded with 1.26 kg/m2 of apple slices. Dry-

ng was run until constant mass was reached.

ctuators B 187 (2013) 234– 240 235

2.2. Color measurements and texture analysis

Quantitative evaluation of the color changes in dried apples wasdone by a portable tri-stimulus colorimeter (Minolta Chroma CR-300, Osaka, Japan) for direct measurement of the color parametersof the raw and dried samples. The color values were expressed as L*(whiteness or brightness/darkness), a* (redness/greenness) and b*(yellowness/blueness) parameters. The diameter of the measuringarea was 8 mm. Measurements were taken at fifteen random pointsfor each sample. The total color difference �E was calculated usingfollowing equation:

�E =√

(�L)2 + (�a)2 + (�b)2 (1)

where �L, �a, �b represent the deviations of the individual valuesfrom the respective values from a fresh raw apple [8].

The study of mechanical properties was carried out using TA-TX2 texture analyzer (Stable Micro Systems Ltd., Surrey, UK) witha 25 kg load cell. Experiment was run with a cutting blade with alength of 62 mm, a width of 24 mm and a thickness of 0.5 mm. Testwas performed to completely cut slice of dried apples with the headvelocity of 1.0 mm/s and with a force of 15 N. For the examination,randomly selected 10 slices of fresh or dried material were used.Work of cutting was calculated as the area under the curve illus-trating the change of force (N) as a function of head shifts (mm) toachieve maximum strength.

2.3. Organoleptic assessment

Sensory evaluation was conducted after 24 h from the dryingprocess. The assessment was performed on a scale from one to fiveby 6-person panel of experts. Dried samples were provided in ran-dom order in transparent containers to the evaluators. To assess thesmell, approximately 50 g of dried material in the bag opened justbefore the evaluation was given to the panelists. For quantitativeexpression of the quality and sensory intensity of dried material,the following factors were used: color, smell, taste, hardness andoverall quality of the dried material. The evaluation was performedtwice in three months apart.

2.4. Preparation of dried apple extracts

Each dried sample was minced using flay. Apple powder wasmixed with distilled water in the same amount which was removedduring the drying process. After 15 min samples were centrifugedat 11,000 rpm for 10 min. Then they were decanted from above thesurface and then passed through a filter paper in order to achievehomogeneous solutions of dried apple extracts.

2.5. Preparation of potentiometric sensors

All inorganic salts used, 1-morpholinoethanesulfonic acid(MES), sulfuric acid and sodium hydroxide were of analyticalgrade and were obtained from Fluka. The working solutionsof salts (0.1 M) were prepared in redistilled water. The poly-meric membrane components: high-molecular weight poly(vinylchloride) (PVC); plasticizers: bis(2-ethylhexyl) sebacate (DOS),o-nitrophenyl octyl ether (o-NPOE); lipophilic salts: potassiumtetrakis[3,5-bis(trifluoromethyl)phenyl]-borate (KTFPB), potas-sium tetrakis(4-chlorophenyl)borate (KTPClPB), tridodecylmethy-lammonium chloride (TDMAC); ionophores: ETH 129 (calciumionophore II), calix[6]arene-hexaacetic acid hexaethylester 4-tert-

butylcalix [4] arene–tetraacetic acid tetraethyl ester (sodiumionophore X), valinomycin (potassium ionophore I), were pur-chased from Fluka. The membrane components (100 mg in total– appropriate ionophore, 20–50 mol% versus ionophore lipophilic

236 A. Kutyła-Olesiuk et al. / Sensors and Actuators B 187 (2013) 234– 240

Table 1Components used for sensor membranes preparation.

Electrode type Plasticizer Lipophilic salt Ionophore Internal filling/conditioning solution

Ca2+-selective DOS 0.85% KTPClPB 2% ETH 129 0.01/0.001 M CaCl2Na+-selective DOS 0.15% KTPClPB 1.7% sodium ionophore X 0.01/0.001 M NaCl

+

sPpnatiataTtaftA

2

b(teprtsobrss

etetwm0ofqe

(3(cwaddTP

caused an increase of a* value from −4.73 to 5.68. On the otherhand, a* value for microwave-convective dried apples was −1.9,which was more similar to the color of raw apples.

Table 2L*, a*, b* values for apples dried by different methods: CD – convective drying, MCD– microwave-convective drying, IRCD – infrared-convective drying.

Material L* a* b*

Fresh 81.9 ± 1.5 y −5.7 ± 0.4 x 22.7 ± 1.7 x

K -selective DOS 0.8% KTFPB

Cation-selective DOS 1% KTFPB

Anion-selective o-NPOE 3.5% TDMAC

alt, 61 wt% plasticizer, and 31–33 wt% high-molecular-weightVC) were dissolved in 2 ml THF. Then, the cocktails (Table 1) wereoured into glass rings in order to evaporate the solvent. In theext step the membrane discs with a diameter of 3 mm were cut offnd miniaturized ion-selective electrodes (ISEs) of classic architec-ure were constructed. The principle of operation of miniaturizedon-selective electrodes is similar to traditional ISEs, however theirrchitecture was adapted to flow-through measurements in a flow-hrough cell [35,36]. The method of the membranes preparationnd the electrodes conditioning were the same as for standard ISEs.he components of internal filling solution and conditioning solu-ion are listed in Table 1. The constructed ISEs were preconditionedt least 24 h. All measurements were carried out in cells of theollowing type: Ag, AgCl; KCl 1 M|CH3COOLi 1 M |analyzed solu-ion ||ion-selective membrane ||internal electrolyte: AgCl, Ag. Theg/AgCl electrode was used as reference electrode.

.6. Hybrid electronic tongue and data analysis

Extracts prepared from raw apples (Fresh) as well as processedy three drying methods: convective (CD), microwave-convectiveMCD) and infrared-convective (IRCD) have been analyzed. Flowhrough measuring cell containing 10 ion-selective electrodes (2lectrodes of every type) was constructed [35]. The selection of theotentiometric sensors was made on the basis of our previous expe-ience [24,25,27,31,35]. To record the signal and for data acquisitionhe multichannel converter Lawson Labs EMF16 and L-EMF DAQ 3.0oftware was used. Measurements of extracts samples were carriedut under flow conditions – the analyzed solution was pumpedy a peristaltic pump to the measuring cell, where steady-stateesponse was observed after few seconds. After reaching steady-tate responses, ten measurement points were collected for eachensor in intervals of 5 s.

During the voltammetric measurements standard 3-electrodelectrochemical cells were applied: classical gold disc working elec-rode, Ag/AgCl reference electrode, and Pt auxiliary electrode. Theselectrochemical measurements were conducted with a 1030 B Elec-rochemical Workstation (CH Instruments, USA). All experimentsere performed in bulk solution at room temperature. The voltam-ograms were registered in the potential range from −0.3 V to

.9 V. The cyclic voltammetry (CV) was conducted at a sweep ratef 100 mV s−1, while the square wave voltammetry (SWV) was per-ormed at a pulse amplitude of 25 mV, increment of 4 mV and a fre-uency of 15 Hz. The differential pulse voltammetry (DPV) param-ters were as follows: pulse amplitude of 50 mV, pulse width 0.2 s.

Amperometric measurements using Glucose Oxidase Biosensorwww.bvt.cz) were carried out at constant potential 0.65 V during00 s. UV–vis spectra were recorded at a wavelength 300–1100 nmLambda 25 UV-Vis spectrometer, PerkinElmer, USA). However, forhemometric analysis only a fragment of the data (300–360 nm)as used, because only in this range significant differences in

bsorbance values were observed for samples prepared by various

rying techniques. For the measurements standard laboratory con-uctometer and pH electrode (pH-meter: Metler Delta 350, Metleroledo, Switzerland; pH electrode: Cole Parmer EW-5991-61, Colearmer, USA) have also been used.

2% valinomycin 0.01/0.001 M KCl– 0.01/0.001 M KCl– 0.01/0.001 M NaCl

All calculations and data analysis were performed in MatLab(The MathWorks, Inc., Natick, USA) and Origin (Microcal Software,Inc, Northampton, USA) software. Chemical images of samples wereprocessed using Partial Least Squares (PLS). Autoscaling as prepro-cessing method and SIMPLS regression algorithm were applied. Forcross-validation, venetian blinds (10 splits) were used. Analysis ofvariance (ANOVA) was conducted to determine the effect of dry-ing on color parameters using Statgraphics software (Statgraphics,Warrenton, Virginia, USA). In order to determine which means aresignificantly different from each other, Duncan test method wasused.

3. Results and discussion

3.1. Color and texture of the dried apples

The consumer associates food color with good processingand safety [12]. It was observed, that the color of convectivedried apple was significantly darker in comparison to raw tis-sue. Moreover, the a* and b* value increased (Table 2). Changesin the color of convective dried tissue were associated with long-lasting high-temperature during drying process [37]. The useof microwave-convective drying caused an increase of L* value.Microwave-convective dried tissue was characterized by a highervalue of the parameter L* in comparison to raw apple. This phe-nomenon results from the way of the assay, which involvesmeasuring the radiation reflected from the surface. Fresh apple con-tains a lot of water and therefore the light reflected from its surfacediffers from light reflected from the porous surface in dried tissue.Water can absorb the radiation, resulting in less reflection from thesurface, which is recorded by the measuring instrument. Indeed,the human eye does not receive the impression of lightening mate-rial (similar conclusion was presented by [9], who obtain respectiveresults for convective and microwave-convective dried garlic).

However, the use of infrared-convective drying of apple tissuecaused darkening. A similar phenomenon was observed by [38].During infrared drying new brown compounds were formed, whichinfluenced the lowering of the brightness of the samples [38]. More-over, in convective and infrared-convective drying also statisticallysignificant change in the a* value was observed, which indicates thebrowning reaction during drying [11]. Similar results was obtainedby Mandala et al. [39], where convective drying of apples at 55 ◦C

CD 77.7 ± 1.2 z 4.2 ± 0.9 z 28.8 ± 0.7 zMCD 86.0 ± 1.7 x −1.9 ± 1.5 y 25.0 ± 0.7 yIRCD 78.2 ± 1.1 z 4.7 ± 1.0 z 30.4 ± 1.1 z

x, y, z: the same letters show homogeneous groups.

A. Kutyła-Olesiuk et al. / Sensors and Actuators B 187 (2013) 234– 240 237

Table 3Characteristics of textural properties for apples dried by different methods: CD –convective drying, MCD – microwave-convective drying, IRCD – infrared-convectivedrying.

Material Water content (%) Max. force (N) Work (mJ)

Fresh 85.8 ± 0.5 v 20 ± 3 v 27 ± 7 vCD 5.7 ± 0.1 y 160 ± 17 x 176 ± 20 xMCD 7.5 ± 0.6 x 240 ± 33 z 310 ± 72 zIRCD 6.0 ± 0.2 x 190 ± 27 y 254 ± 67 y

v

itc(codtmst(

watowop

fatpmtwwcwsead

Fd

, x, y, z: the same letters show homogeneous groups.

During drying, regardless of the method, a significant increasen the b* value was noticed. Dried material obtained withhe use of convective and infrared-convective method washaracterized by significantly higher increase of parameter b*24% and 31%, respectively, Table 2), whereas for microwave-onvective dried apples about 7% increase of this parameterccurred. Basing on the L*a*b* color parameters, the total colorifferences �E were calculated (see equation 1 in Experimen-al section). The total color difference �E was equal 1.8 for

icrowave-convective dried tissue. The second group, which wasignificantly different, was formed by material obtained withhe use of convective (�E = 10.9) and infrared-convective method�E = 10.6).

During the drying many physicochemical processes occur –ater loss and shrinkage of the material, which affects the hardness

nd mechanical properties of tissue are observed [17]. The cuttingest carried out with the dried apple slices showed, that the methodf drying has an impact on the maximum cutting force and work,hich was done to cut dried apple slices (Table 3). Also [40] found

ut, that the drying method significantly influenced the mechanicalroperties of dried apples.

Fig. 1 shows the texture changes of dried apples obtained by dif-erent methods of drying. According to [18], the water content hasn impact on loss of the fragility of foods. Work, which was neededo cut material, and maximum force obtained during cutting wereroportional to the water content in dried tissue. Moreover, in driedaterial with a lower water content and rigid structure, while cut-

ing, cracking inside the material occurred. This fact was connectedith the lowering of maximum force as a result of sample cracking,hich was observed on cutting curves (Fig. 1). In the beginning of

utting in the case of convective dried apple, small cracking peaksere observed. In the case of microwave and infrared dried apple

lices, containing higher water content and the nature of visco-lastic bodies, cracking peaks were not noticed; moreover, worknd maximum force had larger values in comparison to convective

ried material.

ig. 1. Cutting curves of dried apples obtained by different methods: CD – convectiverying, MCD – microwave-convective drying, IRCD – infrared-convective drying.

Fig. 2. Sensory evaluation of apples dried by different methods: CD – convectivedrying, MCD – microwave-convective drying, IRCD – infrared-convective drying.

3.2. Organoleptic assessment

Consumers, when choosing a product, calls attention to itsvisual features, forming a general impression about it [9]. Follow-ing acceptance of the foodstuff is also based on taste sensations.Results of the organoleptic assessment of sensory characteristicsof dried apples was shown in Fig. 2. Based on the overall qual-ity of dried apples, it can be concluded, that the highest markswere received by infrared-convective and convective dried tis-sue, which rated at 4.2 and 4.1 points, respectively. On the otherhand, microwave-convective dried material obtained lower score– 3.4 points. Assessment of the color of dried apples, whichwere obtained using three different methods showed, that thehue and brightness of the samples were similar. Only applesdried with infrared radiation were characterized by slightly darkeryellow color. This could be a result of the formation of brown-colored compounds during infrared-convective drying [38], whichwas also confirmed by instrumental measurement of the bright-ness.

The smell of the dried samples was assessed in relation to thesmell of ripe apples. All samples were characterized by typicalaroma of apples. Strange odors were not noticeable. However, themost intense odor was observed in the case of infrared-convectivedried material, which was assessed at 4.3 points. Similar results wasobtain by [41] during four-hour infrared drying of apples, wherearomas content was unchanged, while convective and microwave-convective dried material was characterized by significantly lessintense flavor.

During eating thickness and rubbery of the dried apples wereassessed. The convective dried samples were rated as the productswith the highest thickness, which probably was related to lowerwater content in the final product. The higher water content inmicrowave-convective and infrared-convective dried apples prob-ably caused changes of samples into rubbery body. However, it isconceivable that the way of providing energy to dry material hadan impact on rubbery state of samples dried using infrared andmicrowave power.

Regardless of the method of drying, the taste of dried apples

was typical and rated at 4.3–4.7 points. However, drying usingmicrowaves causes changes in the intensity of taste, which wasestimated at 3.6 points, whereas the other two methods of drying

2 and Actuators B 187 (2013) 234– 240

lp

3

trmIp(tccb

mct

-

-

aSaofcmdaEpocnoibspcm(eopb

%

ugctaiis

38 A. Kutyła-Olesiuk et al. / Sensors

eads to obtain material with greater intensity of taste (4.4–4.6oints). Strange taste was not noticeable in all samples (Fig. 2).

.3. Electronic tongue analysis

Before the measurements with the electronic tongue system,he verification of sensor performances was carried out. Dynamicesponse curves were recorded, calibration curves were deter-ined, and selectivity coefficients (log K) were calculated for all

SEs used in the sensor array. Slopes, E0 and log K values were appro-riate, showing satisfactory performance of all of the electrodesTable 4). That indicated that they can be used to form the poten-iometric sensor array. The performance of gold disc electrode washecked in ferrocyanide solution and it was correct. The calibrationurves of Glucose Biosensors were consistent with the data giveny the manufacturer.

The prepared electronic tongue was applied to the measure-ents of extracts of dried apples. The recorded data were used to

reate chemical images of the samples obtained with the use of twoypes of electronic tongues:

potentiometric ET – data obtained only from potentiometric mea-surements, 32 samples × 10 features;

hybrid ET – data obtained from all measurements performed(potentiometry, voltammetry, amperometry, conductometry andspectrometry techniques), 32 samples × 224 features.

The first step of the study was the investigation of differentiationbility of the sensor array, which was performed by Partial Leastquares-Discriminant Analysis (PLS-DA). In this task qualitativenalysis was conduced in order to discern between various typesf the dried material and fresh apples. Nominal coding was appliedor 4 classes of the products: fresh, convective dried, microwave-onvective dried, infrared-convective dried, i.e. 4-column targetatrix served as an input for PLS analysis. 3D PLS score plots of

ried apples’ chemical images obtained with potentiometric ETnd hybrid ET are shown in Fig. 3. In the case of potentiometricT clear distinction between chemical images of CD and MCD sam-les is observed, whereas the third cluster is formed by partiallyverlapping chemical images of fresh and IRCD samples. However,ombination of the data provided by various measurement tech-iques realized in hybrid ET, led to improved differentiation abilityf the system – grouping of all chemical images of the same types visible, and various types of the samples are linearly separa-le (Fig. 3b). However, the classification ability of the investigatedystems must be confirmed by the recognition of independent sam-les. Therefore in the following study PLS-DA was applied to thelassification of extracts prepared from raw and dried apples. Dataatrix was divided into train set (75% of the data) and test set

25% of the data). With the use of the first one, PLS-DA model wasstablished, and the latter served for the independent validationf system recognition performance. The results of such analysis isresented as percent of correct classifications, which is expressedy the equation:

CC = number of correct classificationsnumber of all classification

× 100 (2)

This value indicates the probability of assignation of partic-lar sample to its appropriate class. If calculated for train set, itives the estimation of learning ability of the model. When it isalculated for testing samples, it shows real recognition ability ofhe electronic tongue system. %CC was calculated both for train

nd test sets for potentiometric and hybrid ETs. In both systemst reached 100% in the case of training samples, which means thatdeal learning ability is observed even with sensor array based oningle measurement technique. However, independent validation

Fig. 3. 3D PLS score plots of dried apples’ chemical images obtained with (a) poten-tiometric ET and (b) hybrid ET.

of the potentiometric ET with test set data provided poorer result(75%) than in the case of hybrid ET (100%), which is linked withslight overlapping of chemical images of the samples, leading tomisclassifications. Accordingly, RMSE value in the case of hybrid ETis 6 orders of magnitude lower than in the case of potentiometricET (1.2E−11 and 6.6E−5, respectively). Even though sensitivityis 100% in both cases, specificity rises from 58.3% observed forpotentiometric ET to 95.8% for hybrid ET. Therefore it should beemphasized, that ideal classification results were obtained onlyfor hybrid ET, which shows the need to use such devices whereseparability of chemical images of the samples is hard to achieve.

The second step of the study was quantitive analysis, in whichcolor and texture properties achieved with physical methods, aswell as organoleptic scores estimated by the human panel, werecorrelated with sensor array responses. 3 color factors (L*, a*,b*), 3 texture properties (water content, max. force, work), and 9organoleptic scores (overall quality, hue, lightness, typical flavor,flavor intensity, thickness, rubbery, typical taste, taste intensity)were used as features to be estimated with potentiometric and

hybrid ET on the basis of sensor array responses. For each featurePLS model was established. The model performance was char-acterized after linear fitting of the given data (measured withphysical methods or assessed by panelists, placed in the target

A. Kutyła-Olesiuk et al. / Sensors and Actuators B 187 (2013) 234– 240 239

Table 4Working parameters of the miniaturized ISEs.

An electrode type

Ca2+-selective Na+-selective K+-selective Cation-selective Anion-selective

Slope (mV/dec) 27.9 ± 0.8 44.3 ± 0.7 47.7 ± 5.0 16.3 ± 2.8 −55.0 ± 5.9E0 (mV) 116.3 ± 5.9 122.9 ± −32.5 170.7 ± 13.4 113.6 ± 14.8 −130.6 ± 43.1Selectivity coefficients towards

Ca2+ 0.0 −2.6 −3.4 −1.5 –K+ −3.1 −1.9 0.0 0.2 –Li+ −2.4 −2.3 −3.3 −0.3 –Na+ −2.9 0.0 −2.9 0.0 –NH4

+ −3.1 −2.2 −1.5 0.2 –Mg2+ −2.3 −2.2 −3.5 −0.4 –Cl− – – – – −1.6SO4

2− – – – – −2.0ClO4

− – – – – 2.6NO3

− – – – – 0.0

Table 5Parameters of linear fitting of real and PLS-predicted color properties.

Potentiometric ET Hybrid ET

TRAIN TEST TRAIN TEST

L* R2 0.983 0.774 0.999 0.974a 0.98 0.77 1.00 1.02b 1.41 19.63 0.08 −0.75RMSE 9.4E−02 7.9E−01 2.3E−02 3.6E−01

a* R2 0.983 0.930 1.000 0.989a 0.98 0.94 1.00 1.01b 0.01 0.43 0.00 −0.35RMSE 1.2E−01 4.9E−01 1.3E−02 2.4E−01

b* R2 0.977 0.855 1.000 0.986a 0.98 0.93 1.00 1.00

midt0towihwswf

TP

Table 7Parameters of linear fitting of real and PLS-predicted organoleptic scores.

Potentiometric ET Hybrid ET

TRAIN TEST TRAIN TEST

Overall quality R2 0.992 0.663 1.000 0.970a 0.99 0.71 1.00 0.78b 0.03 0.97 0.00 0.87RMSE 7.7E−03 8.9E−02 3.5E−04 3.3E−02

Hue R2 0.999 0.995 1.000 0.990a 1.00 1.13 1.00 1.00b 0.00 −0.28 0.00 0.01RMSE 1.2E−03 1.0E−02 9.4E−05 6.3E−03

Lightness R2 1.000 0.997 1.000 0.975a 1.00 1.00 1.00 0.96b 0.00 0.02 0.00 0.13RMSE 3.8E−04 1.9E−03 6.8E−05 5.4E−03

Typical flavor R2 1.000 0.999 1.000 0.975a 1.00 1.00 1.00 0.96b 0.00 0.01 0.00 0.17RMSE 3.8E−04 1.9E−03 6.8E−05 5.4E−03

Flavor intensity R2 1.000 0.998 1.000 0.967a 1.00 0.91 1.00 0.93b 0.00 0.25 0.00 0.26RMSE 3.0E−03 2.9E−02 4.6E−04 3.8E−02

Thickness R2 0.994 0.885 1.000 0.995a 0.99 1.24 1.00 0.96b 0.02 −1.01 0.00 0.12RMSE 1.8E−02 2.0E−01 3.9E−04 2.6E−02

Rubbery R2 0.993 0.846 1.000 0.992

b 0.60 2.48 0.01 −0.24RMSE 9.7E−02 5.2E−01 1.1E−02 1.7E−01

atrix) to the predicted data (PLS predicted value of every featuren PLS output matrix). The values of the slope (a), intercept (b),etermination coefficient (R2), and RMSE were calculated for therain and test samples. The ideal values of these parameters are 1,, 1, and 0, respectively. Generally, better values were obtained inhe case of train sets and hybrid ET (Tables 5–7). In the estimationf color properties, the worst results were obtained when L* valueas determined (R2 = 0.774, a = 0.77), but they could be easily

mproved when all the data were considered by the model inybrid ET (R2 = 0.974, a = 1.02; Table 5). All texture properties

ere also satisfactory estimated, especially in the case of hybrid

ystem (Table 6). Significant lowering of b and RMSE, occurringhen all measurement techniques were involved, is clearly visible

or both train and test sets. The same trend can be observed for

able 6arameters of linear fitting of real and PLS-predicted texture properties.

potentiometric ET hybrid ET

TRAIN TEST TRAIN TEST

Water content R2 0.977 0.810 1.000 0.999a 0.98 0.73 1.00 1.01b 0.01 0.04 0.00 −0.02RMSE 1.0E−02 6.6E−02 3.5E−05 6.8E−03

Max. force R2 0.926 0.923 0.994 0.984a 0.93 0.48 0.99 1.03b 11.31 100.10 0.92 4.53RMSE 4.7E+00 2.0E+01 1.4E+00 5.7E+00

Work R2 0.908 0.764 0.992 0.977a 0.91 0.41 0.99 1.02b 17.60 143.17 1.57 10.22RMSE 6.9E+00 3.0E+01 2.0E+00 8.6E+00

a 0.99 1.20 1.00 0.94b 0.02 −0.28 0.00 0.17RMSE 2.2E−02 2.5E−01 5.1E−04 4.1E−02

Typical taste R2 0.999 0.996 1.000 0.961a 1.00 0.85 1.00 0.91b 0.00 0.64 0.00 0.40RMSE 1.2E−03 1.3E−02 1.5E−04 1.2E−02

Taste intensity R2 0.998 0.983 1.000 0.955a 1.00 0.77 1.00 0.87

b 0.01 0.87 0.00 0.54RMSE 4.2E−03 4.8E−02 3.8E−04 3.4E−02

9 organoleptic scores (Table 7). It led us to the conclusion, thatthe investigated color and texture properties, as well as tasteand flavor scores, can be determined with the developed hybridelectronic tongue with satisfactory correctness.

4. Conclusions

In this work an electronic tongue based on potentiometry,voltammetry, amperometry, conductometry and spectroscopy was

2 and A

ptstettbtTtds

A

a(tt

R

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

40 A. Kutyła-Olesiuk et al. / Sensors

roposed for the evaluation of organoleptic scores, color and tex-ure properties of dried apple samples. The fusion of various sensingchemes, i.e. the realization of the hybrid electronic tongue, leado the improved recognition ability of the system, which wasvidenced by the increasing number of correct classifications. Inhe case of quantitive analysis, i.e. the determination of organolep-ic, color and texture properties, also better results were providedy the hybrid system – values of a, b, R2, RMSE of linear fitting ofhe real data and PLS-predicted data were closer to the ideal ones.his work proved, that data fusion in hybrid ET allows for a bet-er characterization of the apple extracts, therefore the developedevice could be used for qualitative and quantitive analysis of suchamples.

cknowledgements

This work has been supported by National Center for Researchnd Development within a framework of LIDER programmeNr LIDER/17/202/L-1/09/NCBiR/2010). The authors would like tohank Sylwia Dziedzic-Talipska and Wojciech Jankowski for help inhe measurements.

eferences

[1] S.K. Chou, K.J. Chua, New hybrid drying technologies for heat sensitive food-stuffs, Trends In Food Science & Technology 12 (2001) 359–369.

[2] H. Vega-Mercado, M.M. Angora-Nieto, G.V. Bartosa-Cánovas, Advanced indehydration of food, Journal of Food Engineering 49 (2001) 271–289.

[3] P.P. Lewicki, Design of hot air drying for better foods, Trends In Food Science &Technology 17 (2006) 153–163.

[4] Y. Deng, Y. Zhao, Effect of pulsed vacuum and ultrasound osmopretreatmentson glass transition temperature, texture, microstructure and calcium pene-tration of dried apples (Fuji), LWT – Food Science and Technology 41 (2008)1575–1585.

[5] K.J. Chua, S.K. Chou, Low-cost drying methods for developing countries, TrendsIn Food Science & Technology 14 (2003) 519–528.

[6] C. Ratti, A.S. Mujumdar, Drying of fruit, in: D.M. Barrett, L. Somogyi, H.Ramaswamy (Eds.), Processing Fruit, CRC Press, New York, 2005 (Chapter 7).

[7] D. Witrowa-Rajchert, M. Rzaca, Effect of drying method on the microstructureand physical properties of dried apples, Drying Technology 27 (2009) 903–909.

[8] M. Maskan, Kinetics of colour change of kiwifruits during hot air and microwavedrying, Journal of Food Engineering 48 (2001) 169–175.

[9] G.P. Sharma, S. Prasad, Drying of garlic (Pallium sativum) cloves by microwave-hot air combination, Journal of Food Engineering 50 (2001) 99–105.

10] W. Kalt, Effect of production and processing factors on major fruit and vegetableantioxidants, Journal of Food Engineering 70 (2005) 11–19.

11] R. Vadivambal, D.S. Jayas, Changes in quality of microwave-treated agriculturalproducts – a review, Journal of Biosystems Engineering 98 (2007) 1–16.

12] J. Wilska-Jeszka, Food colorants, in: Z.E. Sikorski (Ed.), Chemical and FunctionalProperties of Food Componants, CRC Press, Boca Raton, 2007, pp. 245–274.

13] G. Sumnu, E. Turabi, M. Oztop, Drying of carrots in microwave and halogenlamp-microwave combination movens, LWT – Food Science and Technology38 (2005) 549–553.

14] K. Krishnamurth, H.K. Khurana, S. Jun, J. Irudayaraj, A. Demirci, Infrared heatingin food processing: an overview, Comprehensive Reviews in Food Science andFood Safety 7 (2008) 2–13.

15] M. Maskan, Microwave/air and microwave finish drying of banana, Journal ofFood Engineering 44 (2000) 71–78.

16] J. Varith, P. Dijkanarukkul, A. Achariyaviriya, S. Achariyaviriya, Combinedmicrowave-hot air drying of peeled longan, Journal of Food Engineering 81(2007) 459–468.

17] P.P. Lewicki, E. Jakubczyk, Effect of hot air temperature on mechanical proper-ties of dried apples, Journal of Food Engineering 64 (2004) 307–314.

18] T. Labuza, K. Roe, C. Payne, F. Panda, T.J. Labuza, P.S. Labuza, L. Krusch, in: M.Silva, S. Rocha (Eds.), Storage Stability of Dry Food Systems: Influence of StateChanges During Drying and Storage Drying 2004, Ourograf Grafica Campinas,Sao Paulo, 2004, pp. 48–68.

19] P. Ciosek, W. Wróblewski, Sensor arrays for liquid sensing – electronic tonguesystems, The Analyst 132 (2007) 963–978.

20] K. Toko, Taste sensor with global selectivity, Materials Science and Engineering

4 (1996) 69–82.

21] P. Ciosek, W. Wróblewski, Potentiometric electronic tongues for foodstuff andbiosample recognition – an overview, Sensors 11 (2011) 4688–4701.

22] A. Gutes, F. Cespendes, M. del Valle, Electronic tongues in flow analysis, Ana-lytica Chimica Acta 600 (2007) 90–96.

ctuators B 187 (2013) 234– 240

23] A. Rudnitskaya, A. Ehlert, A. Legin, Y. Vlasov, S. Buttgenbach, Multisensor systemon the basis of an array of non-specific chemical sensors and artificial neuralnetworks for determination of inorganic pollutants in a model groundwater,Talanta 55 (2001) 425–431.

24] A. Kutyła-Olesiuk, M. Zaborowski, P. Prokaryn, P. Ciosek, Monitoring of beerfermentation based on hybrid electronic tongue, Bioelectrochemistry 87 (2012)104–113.

25] P. Ciosek, K.W. Brudzewski, Wróblewski Milk classification by means of anelectronic tongue and support vector machine neural network, MeasurementScience and Technology 17 (2006) 1379–1384.

26] P. Ciosek, I. Grabowska, Z. Brzózka, W. Wróblewski, Analysis of dialysate fluidswith the use of a potentiometric electronic tongue, Microchimica Acta 163(2008) 139–145.

27] M. Janczyk, A. Kutyła, K. Sollohub, H. Wosicka, K. Cal, P. Ciosek, Electronic tonguefor the detection of taste-masking microencapsulation of active pharmaceuti-cal substances, Bioelectrochemistry 80 (2010) 94–98.

28] F. Winquist, S. Holmin, C. Krantz- Rulcker, P. Wide, I. Lundstrom, A hybridelectronic tongue, Analytica Chimica Acta 406 (2000) 147–157.

29] A. Legin, A. Rudnitskaya, Y. Vlasov, C. Di Natale, F. Davide, A. D’Amico, Tastingof beverages using an electronic tongue, Sensors and Actuators B: Chemical 44(1997) 291–296.

30] R. Labrador, J. Soto, R. Martınez-Manez, L. Gil, An electronic tongue for quali-tative and quantitative analyses of anions in natural waters, Journal of AppliedElectrochemistry 39 (2008) 2505–2511.

31] P. Ciosek, E. Augustyniak, W. Wróblewski, Polymeric membrane ion-selectiveand cross-sensitive electrode-based electronic tongue for qualitative analysisof beverages, The Analyst 129 (2004) 639–644.

32] Z. Kovács, L. Sipos, D. Szöllõsi, Z. Kókai, G. Székely, A. Fekete, Electronic tongueand sensory evaluation for sensing apple juice taste attributes, Sensor Letters9 (2011) 1273–1281.

33] A. Rudnitskaya, A. Legin, S. Makarychev-Mikhailov, O. Goryacheva, Y. Vlasov,Quality monitoring of fruit juices using an electronic tongue, Analytical Sci-ences 17 (Suppl.) (2001) i309–i312.

34] A. Rudnitskaya, D. Kirsanov, A. Legin, K. Beullens, J. Lammertyn, B.M. Nicolaı, J.Irudayaraj, Analysis of apples varieties – comparison of electronic tongue withdifferent analytical techniques, Sensors and Actuators B: Chemical 116 (2006)23–28.

35] E. Witkowska, A. Buczkowska, A. Zamojska, K.W. Szewczyk, P. Ciosek, Moni-toring of periodic anaerobic digestion with flow-through array of miniaturizedion-selective electrodes, Bioelectrochemistry 80 (2010) 87–93.

36] J. Markiewicz, J. Taff, B. Wysocki, W. Wróblewski, Modular flow-through cell ofthe ion-selective electrode for water quality monitoring, Date of application:21.07.2008 P384364.

37] G. Mazza, Health aspects of natural colors, in: G.J. Lauro, F.J. Francis (Eds.),Natural Colors, Marcel Dekker, Inc., New York, 2000, pp. 289–314.

38] D. Nowak, P.P. Lewicki, Infrared drying of apple slices, Innovative Food Scienceand Emerging Technologies 5 (2004) 353–360.

39] I.G. Mandala, E.F. Anagrostaras, C.K. Oikonmou, Influence of osmotic dehydra-tion conditions on apple air-drying kinetics and their quality characteristics,Journal of Food Engineering 69 (2005) 307–316.

40] N. Leeratanarak, S. Devahastin, N. Chiewchan, Drying kinetics and quality ofpotato chips undergoing different drying technics, Journal of Food Engineering77 (2006) 635–643.

41] S. Timoumi, D. Mihoubi, F. Zagrouba, Schrinkage, vitamin C degradation andaroma losses during infra-red drying of apple slices, LWT – Food Science andTechnology 40 (2007) 1648–1654.

Biographies

Anna Kutyła-Olesiuk obtained her M.Sc. Chemistry from Warsaw University ofTechnology. Now, she is a Ph.D. student in Department of Microbioanalytics, WUT.Her work involves design and construction of electrochemical sensor arrays for therecognition and classification of biological samples.

Małgorzata Nowacka graduated in Warsaw University of Life Sciences at Facultyof Food Science, Department of Food Engineering and Process Management (2005).On 11/12/2009, she received a Ph.D. in agricultural sciences in the field of food tech-nology and nutrition at this University. From 30 December 2009 she is an assistantprofessor in the Department of Food Engineering and Process Management. Sheis the author and co-author of 17 original papers, 13 popular science publicationsand 2 books chapter. Moreover she is also a reviewer of articles in national andinternational journals.

Małgorzata Wesoły received engineering degree from Warsaw University of Tech-nology. Now, she continues her studies in Department of Microbioanalytics, WUT.

Patrycja Ciosek received M.Sc. degree in chemistry from Warsaw University ofTechnology (WUT, Poland), and Ph.D. in analytical chemistry at the same Univer-sity (2006). Since then she has been working as assistant professor at WUT. In herresearch she is focused on electronic tongue, multisensor systems, and applicationsof various numeric methods to analyze sensor array responses.