Journal of Food Engineering 116 (2013) 50–56

Contents lists available at SciVerse ScienceDirect

Journal of Food Engineering

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

Minimally-destructive evaluation of durian maturity based on electricalimpedance measurement

Pramote Kuson a,b, Anupun Terdwongworakul a,⇑a Department of Agricultural Engineering, Faculty of Engineering at Kamphaengsaen, Kasetsart University, Kamphaengsaen, Nakhon Pathom, Thailandb The Postharvest Technology Innovation Center (PHTIC), Higher Education Commission, Bangkok 10400, Thailand

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

Article history:Received 29 May 2012Received in revised form 23 November 2012Accepted 26 November 2012Available online 5 December 2012

Keywords:DurianImpedance spectroscopyPartial least squares regressionDiscriminant analysis

0260-8774/$ - see front matter � 2012 Elsevier Ltd. Ahttp://dx.doi.org/10.1016/j.jfoodeng.2012.11.021

⇑ Corresponding author. Tel.: +66 86 7751723; fax:E-mail address: anupunterd@yahoo.com (A. Terdw

Electrical impedance spectroscopy was investigated to model the dry matter content of durian using par-tial least squares regression. Measurements of the impedance were taken on the stem and the rind of dur-ian samples at various stages of maturity based on the number of days after anthesis. Plots of therelationship between resistance and reactance, and the change in impedance and capacitance withrespect to frequency in a range of 1–200 kHz were explored to determine the optimal frequencies asso-ciated with variation in the number of days after anthesis. The impedance parameters at frequencies of 1,41 and 200 kHz were employed to model the dry matter content of the pulp. The reactance of the crosssection of the stem and the capacitance of the rind were found to predominantly contribute to the pre-diction of the dry matter content. Selected impedance parameters using a stepwise regression could beused to classify durian samples into an immature class and mature class with less accuracy of 83.3%.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Durian is regarded as one of the more favored tropical fruits(Siriphanich, 2011). Export of durian from Thailand is the largestproportion of production (Cunningham, 2000). Inclusion of imma-ture durian is the major problem for the export market as imma-ture durian provides unacceptable quality. At present, harvestingof physiologically mature durian is undertaken by manual practice.The maturity indices commonly used are the number of days afteranthesis, sound pitch when tapping, resistance of the spines whenpressed against each other, bending strength of the stem, sweet-ness of the solution from the cut stem and other procedures. Theseharvesting indices are prone to error caused by the inexperience orfatigue of the person harvesting.

In efforts to replace manual harvesting techniques, a number ofobjective techniques have been studied for maturity evaluation.The non-destructive techniques that have been investigated werebased on subjective indices employed by the harvesters. Measure-ments of frequency response to forced vibration and ultrasonic re-sponse carried out in the region between the spines at the middlepart of the durian (Kongrattanaprasert et al., 2001) had an accuracyrate of 90% for maturity determination. A microwave techniquebased on the reflection coefficient at 3 GHz (Rutpralom et al.,2002) was also reported as a promising method for determinationof durian maturity. The application of the acoustic response to freevibration was also studied for the maturity detection of durian

ll rights reserved.

+66 34 351896.ongworakul).

(Neamsorn and Terdwongworakul, 2004). Nevertheless, the men-tioned techniques have yet to be implemented on the productionline.

Electrical impedance spectroscopy (EIS) has been of interest as atechnique for assessment of the in vivo condition of plant tissues.EIS is considered to be a rapid and easy technique for measure-ment. With EIS, the structure of plant cells can be modeled as anelectrical circuit diagram (Zhang et al., 1990). This enables inter-pretation of the impedance spectra acquired by EIS in terms ofcomponents of resistance and capacitance with respect to the cel-lular structures. For example, the extent of tissue damage of applebruise was successfully detected (Jackson and Harker, 2000). EISwas also investigated as a technique to assess ripening of fruit(Harker and Dunlop, 1994; Harker and Maindonald, 1994). For nec-tarines for instance, a decrease in the resistance of the apoplastwas observed immediately after harvest (Harker and Maindonald,1994). A non-destructive impedance spectroscopic technique waspreliminarily investigated for future development of a robotic har-vesting arm for mango (Rehman et al., 2011). Impedance spec-trometry was shown to be capable of characterizing raw and ripemangos in the frequency range of 1–200 kHz. For durian fruit,few reports were found on the application of electrical propertiesfor evaluation of durian maturity.

However, some non-destructive techniques that offer directmeasurement of pulp content are not practical for use with durianbecause of its large size and thick husk. As mentioned previously, anumber of the maturity indices which are used in practice are asso-ciated with the stem and the rind. As such, indirect measurementof the pulp content based on minimally-destructive measurement

P. Kuson, A. Terdwongworakul / Journal of Food Engineering 116 (2013) 50–56 51

of the stem and rind appeared to be viable for assessment of thefruit’s maturity.

The current research reported on the application of EIS for pre-diction of the pulp dry matter in association with maturity. Opti-mal parameters derived from EIS of the rind and the stem, whichvaried with maturity, were explored for modeling maturity. Classi-fication of durian into the mature and the immature stages pre-grouped on a basis of the dry matter content was also investigated.

2. Materials and methods

2.1. Sample preparation

Durians of the ‘Monthong’ variety used in the study were har-vested from trees in a plantation in Chanthaburi, Thailand. Thestudy was carried out during March and May 2010. Flower taggingstarted after anthesis and the fruits were randomly harvested atthe chronological stages of 117, 124, 131 and 138 days after anthe-sis (DAA). These stages were chosen to cover the variation of matu-rity from immature to over-mature levels. At each harvest, 30fruits were obtained and transported immediately to the labora-tory where they were acclimatized at a controlled temperature of25 �C and a relative humidity of approximately 85% for 24 h priorto subsequent measurements.

2.2. Measurement of electrical impedance spectroscopy

The EIS measuring system used in the study is shown in Fig. 1.To acquire the EIS parameters, an available pincher probe (HIOKI9143, Japan) with two parallel electrode tips was inserted intothe stem and the rind to a depth of 3 mm (full length of electrodetip) with the tips spaced 10 mm apart. This kind of probe measuresthe sample impedance based on a two-terminal configuration.Although, this configuration is most commonly used for biologicaltissue investigation, the measured impedance normally includeselectrode polarization impedance (Ackmann and Seitz, 1984). Theelectrode polarization impedance can be much larger than thesample impedance especially at frequencies below 1 kHz (Harkerand Dunlop, 1994; Ferris, 1974). To minimize the electrode polar-ization impedance, the impedance data in a frequency range of 1–200 kHz at an increment of 2 kHz and a measuring voltage of 1 Vwere recorded using an LCR meter (HIOKI 3532-50, Japan).

Biological tissue can be represented by a simplified electricalmodel consisting of a parallel circuit of a capacitor and a resistor(Gourshete et al., 2005). Additionally, in preliminary tests on sam-ples, relatively high impedance in the range of 1000–8000 ohmsand low capacitance from 1 to 4 nF were observed. Hence, a paral-lel equivalent circuit mode was selected for proper measurement,which was suitable for test samples with high impedance andlow capacitance, as suggested in the instruction manual. All re-corded data were transferred to a computer for further analysesby the software HIOKI LCR-RS232C V4.01e (Nagano, Japan). The

Stem

LCR meter

Durian

Pincher probe

Computer

Fig. 1. Layout of the instrument for measurement of the electrical impedance onthe stem of durian sample.

parameters measured by the LCR meter comprised capacitance(C), resistance (R), reactance (X) and impedance (Z).

2.2.1. EIS measurement of the stemDurian stem normally is divided into two parts by an abscission

zone (Fig. 2). The upper part of the stem is attached to the tree andthe lower part is connected to the fruit. Measurement of the EISparameters was conducted by inserting the two tips of the pincherprobe at the center of the lower part (LP) of the stem on the axisline (Fig. 2a). Three repetitions of measurements 120� apart weremade around the stem. The upper part was then removed andthe subsequent measurement of the EIS parameters was carriedout on the cross sectional surface of the lower part of the stem(CS) as indicated in Fig. 2b. The two tips of the probe were insertedat the perimeter of a fixed diameter circle of 10 mm. The measure-ment was repeated three times on the perimeter with each mea-surement 120� apart. The average of the three measurementswas used for subsequent statistical analyses.

2.2.2. EIS measurement on the rindNormally, each fruit contains five locules. According to durian

standard, fruit should have a minimum of three fertile locules, whichwere the locules that appeared fully filled along the length of thefruit (National Bureau of Agricultural Commodity and Food Stan-dards, 2003). Three fertile locules of each fruit were thus selectedfor the EIS measurement. Selection of the three locules to representthe whole fruit followed previous investigation (Siriphanich andJerapat, 2005). The tips of the probe were inserted along the longitu-dinal suture line of the locule. Each locule was measured at the mid-dle part of the rind (RI). A total of three measurements from eachfruit were averaged and applied for further statistical analyses.

2.3. Determination of dry matter

During growth and maturation of ‘Monthong’ durian, dry mat-ter of the durian pulp was proved to be a suitable and the best in-dex of maturity (Sangwanangkul and Siriphanich, 2000). As such,the dry matter of the pulp was referenced as a standard index ofmaturity. Determination of the dry matter involved taking pulpfrom the middle segments of every locules, cutting it into smallpieces, and thoroughly mixing. A sample of about 20 g of the mixedpulp was dried in a hot air oven at 70 �C for 48 h (Sangwanangkuland Siriphanich, 2000). The dry matter percentage (%DM) was ex-pressed as (w2/w1) � 100 where w1 is the original mass of the pulp(g) and w2 is the dried mass (g).

2.4. Statistical analyses

Partial least squares regression (PLSR) analysis was performedto build a model to predict the dry matter content of the pulp.

(a)Stem cross section

Lower stem 10 mm

Abscission zone

(b)

Fig. 2. Positions of measurements indicated by solid dots on (a) the stem and (b)the stem cross section using the pincher probe.

52 P. Kuson, A. Terdwongworakul / Journal of Food Engineering 116 (2013) 50–56

The samples in each DAA group were sorted based on the dry mat-ter value and alternately selected for sub-calibration and sub-pre-diction sets. As such, both sub-sets covered identical variations anduniform distribution of the percentage of dry matter. With all cor-responding sub-sets in DAA groups of durian combined together,the calibration set contained 90 samples and the prediction set30 samples (Table 1). The calibration set was used to build a cali-bration model with the application of PLSR in the UnscramblerV9.8 program (Camo, Oslo, Norway). The optimal number of PLSRfactors was determined by full cross validation with the loweststandard error of cross validation based on the calibration sample.

Models were created using EIS parameters as independent vari-ables selected at useful frequencies from plots of the reactanceagainst the resistance and the change in the impedance as wellas the capacitance with frequency. The useful frequencies, at whichthe effect of maturity on the parameters in the above-mentionedplots was observed, were from 1 kHz to 200 kHz. In the searchfor the optimal EIS parameters, stepwise regression was performedwith all selected EIS parameters using the statistical analysis soft-ware (SPSS version 9.0, SPSS Inc., Chicago, USA) to model the drymatter content. In the process, the stepwise method chooses thesignificant predictor of a dependent variable entered in each stepbased on the statistical criteria. The chosen predictor is the onethat provides a statistically significant regression coefficient. Themethod continues by evaluating the chosen predictor as well asnewly entered predictors. The chosen predictor is removed if it isno longer significant. The procedure is repeated until there arenot more significant predictors to enter and remove.

The prediction performance of the model was validated by com-parison of the measured values and the model prediction values ofthe samples in the prediction set. The statistics used for the perfor-mance assessment were the correlation coefficient (rp), root meansquared error of prediction (RMSEP) and bias.

For the prospective application to sort the immature durians inthe sample, a classification model was also created using discrim-inant analysis on the basis of the EIS parameters selected by thestepwise regression. Normally, ‘Monthong’ durian fruit is judgedmature when the dry matter reaches around 32% (National Bureauof Agricultural Commodity and Food Standards, 2003). The duriansamples were re-divided into two groups based on the dry mattercontent. For the purpose of practical sorting, samples with drymatter content lower than 32% were assigned to the immaturegroup and the remaining samples were allocated to the maturegroup. In the immature group, the samples were sorted in ascend-ing order with respect to the dry matter value. The samples werethen sub-divided by alternatively selecting the samples into asub-calibration set (75%) and sub-prediction set (25%). Thisprovided a similar distribution of the dry matter content of thesamples in the sub-calibration and sub-prediction sets. Similarly,samples in the mature group were sub-divided into a sub-calibration set and sub-prediction set. These corresponding sub-sets were then combined into the calibration set (90 samples)and prediction set (30 samples). The calibration set of the mature

Table 1Statistics of dry matter content of durian samples grouped based on the calibration and preanthesis.

Grouped for PLSR

Calibration set Prediction set

Maximum value 42.5% 42.2%Minimum value 10.3% 11.9%Mean value 32.7% 33.1%Standard deviation 7.8% 7.6%Number of samples 90 30

* Different letters in a row indicate a significant difference at p < 0.05.

and the immature groups was used to develop the classificationmodel using discriminant analysis (SPSS version 9.0, SPSS Inc., Chi-cago, USA). The performance of the model was validated using thesamples in the prediction sets for the correctly classified samplesinto the mature and the immature groups.

3. Results and discussion

3.1. Effect of maturity on dry matter content

The average dry matter content was significantly affected bymaturity at p < 0.05 (Table 1). The increase of 47.8% in the dry mat-ter from the immature fruit (22.6% at 117 DAA) to mature fruit(33.4% at 124 DAA) was greater than that from mature fruit toover-mature fruit at 131 DAA (8.2% increase). The result coincidedwith the report by Sangwanangkul and Siriphanich (2000).

3.2. Determination of optimal frequencies of the EIS parameters

Plots of change in the impedance and capacitance with fre-quency and the reactance against the resistance were explored todetermine the frequencies at which an apparent difference in theEIS parameters was observed with respect to the variation in stageof maturity.

3.2.1. Change in impedance with respect to frequencyThe plot of impedance (Z) with respect to the logarithm of fre-

quency was explored. For biological systems, the cell membranecan be represented by an equivalent circuit consisting of a capaci-tor and a resistor in parallel connection with the intercellular andextracellular solutions as two additional series resistors (Angers-bach et al., 1999). Generally, the electrical impedance of biologicaltissue reduces with an increase in frequency because of capacitivecharacteristics of the cell membrane. In the case of the stem andthe rind, a drop in Z was also observed as the frequency was in-creased (Fig. 3). A similar decline of the Z value was also reportedwith kiwifruit (Sugiyama et al., 1987).

For the stem, the Z value was continually lower as the fruitdeveloped from the immature stage to the over-mature stage with-in a range of measured frequencies. Chattavongsin and Siriphanich(1987) studied the anatomy of durian stem at various stages ofmaturity. They found that as the Monthong durians matured, thenumber of cortical sclereids in the outer part of the lower stem in-creased. The sclereid normally develops a secondary wall, which isusually lignified making it very strong and impermeable to water(Beck, 2010). It was speculated that as fruits matured, the sclereidreleased intracellular electrolytes into the extracellular liquid untilthe cells were dead. Thus, an increase in the sclereid resulted inmore electrolytes in the extracellular liquid and a decrease inimpedance. The Z values of the lower stem (Fig. 3a) were higherthan those of the stem cross section (Fig. 3b) and the rind(Fig. 3c). On average, measurement at the stem cross section

diction sets for partial least squares regression (PLSR), and on the number of days after

Grouped based on days after anthesis

117 124 131 138

31.1% 40.6% 41.7% 42.5%10.3% 23.8% 29.6% 32.6%22.6%a* 33.4%b 36.4%c 39.4%d

5.9% 4.6% 3.5% 2.3%30 30 30 30

Fig. 3. Change in impedance with respect to frequency of (a) lower stem, (b) stemcross section and (c) rind.

P. Kuson, A. Terdwongworakul / Journal of Food Engineering 116 (2013) 50–56 53

obtained the lowest value of the Z value. In the stem cross section,fibers in xylem were aligned in radial orientation (Chattavongsinand Siriphanich, 1987). Thus, this form of organization of fiberscreated more available extracellular pathways (low impedance)to the radially applied electric field. On the other hand, the lowerstem and rind were subjected to a longitudinal electric field. Themonotonous decrease in the Z value was obvious specifically inthe case of the stem cross section.

Regarding the rind, a consistent change in the Z value with re-spect to the DAA value was not apparent (Fig. 3c). However, theimpedances of the 117 DAA and 124 DAA were close and lowerthan those of the 131 DAA and 138 DAA. This showed that rindimpedance had a tendency to rise with greater differences in matu-rity. According to Koksungnoen and Siriphanich (2005) when dur-ian enter the mature stage, cell size in the rind remains constant.However, tannin in the cell diminished continuously. This probablycaused a higher concentration of the tannin in the extracellular

liquid. There was a report that tannin could be used as a steel cor-rosion inhibitor (Peres et al., 2012). This implies that a high con-centration of tannin in the given electrolyte solution providedhigh impedance. Therefore, it could be argued that more maturefruit have higher impedance in the rind than less mature ones.The divergence of the Z value with respect to DAA was highest atthe lowest frequency (1 kHz) in the measurement.

3.2.2. Change in capacitance with respect to frequencyGenerally, capacitance (C) exhibited a sharp drop in value up to

around 5 kHz and then the rate of decrease was slow and started tolevel off towards a high frequency. The pattern of decrease in Ccoincides with a report by Rehman et al. (2011) for mangos. In or-der to more clearly examine the change in C, the values of C from 1to 5 kHz pertaining to the rapid drop were explored (Fig. 4).

An increase in the C value was indicated with respect to DAA. Inthe case of the lower stem, lignified sclereid has a thick wall andthus a high capacitance (Chattavongsin and Siriphanich, 1987).As a result, more mature fruit generally demonstrated highercapacitance in the lower stem due to the increasing number ofsclereids.

Additionally, the capacitance of the lower stem increased from117 DAA to 124 DAA but decreased at 131 DAA before rising againat 138 DAA (Fig. 4a). This inconsistent increase may be explainedby the fact that the dry matter range of 124 DAA (23.8% and40.6% in Table 1) overlapped the corresponding range of 131 DAA(29.6% and 41.7% in Table 1) to a larger extent than other ranges(Table 1). In addition, the variation in the capacitance of the lowerstem with respect to DAA (Fig. 4a) was smaller than the stem crosssection (Fig. 4b). Thus, it was possible that the capacitances of sam-ples from 124 DAA to 131 DAA were similar and did not show anytrend.

The C value of the stem cross section obtained a monotonous in-crease with the DAA value (Fig. 4b). In addition, variation of theaverage C value of the stem cross section was greatest comparedto the lower stem (Fig. 4a) and the rind (Fig. 4c). This could be ex-plained by the fact that the stem cross section was measured radi-ally and there were different types of cells with various shapes andorientations in radial arrangement (Chattavongsin and Siriphanich,1987). Therefore the different types of cells tended to restrict con-tinuous electrolyte path-ways and as such, caused the highestcapacitance.

For the rind, no difference in the C value with respect to the DAAvalue was apparent (Fig. 4c). This could possibly be explained bythe report that the size of the cell was relatively unchanged whenthe durian reached full maturity (Koksungnoen and Siriphanich,2005).

Overall, the divergence of the C value was clear at a frequency of1 kHz

3.2.3. Empirical selection of the optimal frequency from plots betweenresistance and reactance

To reduce the number of electrical parameters for analysis, opti-mal frequencies were empirically selected from plots that showeda relationship between resistance and reactance. The optimal fre-quency was empirically defined as the frequency at which differ-ence in resistance or reactance was affected by maturity.

In general, impedance data are composed of resistance andreactance components (Stout, 1988). The reactance of the biologi-cal system normally quantifies the capacitance of biological mem-branes (Harker and Maindonald, 1994). Variation in electricalimpedance data was expressed in a plot of reactance against resis-tance at each frequency from 1 to 200 kHz with respect to DAA asshown in Fig. 5. A continuous decrease in the resistance was ob-served as the frequency was increased (right to left in Fig. 5). Thereactance value of all DAA groups rose to a maximum in the

Fig. 4. Change in capacitance with respect to frequency of (a) lower stem, (b) stemcross section and (c) rind.

400

600

800

1000

1200

1400

1600

1800

2000

2200

1500 3500 5500 7500

Rea

ctan

ce(o

hm)

Resistance (ohm)

117 DAA

124 DAA

131 DAA

138 DAA

1 kHz

200 kHz

41 kHz

(a) Lower stem

Fig. 5. Relationship between reactance and resistance with respect to frequency of(a) lower stem, (b) stem cross section and (c) rind.

54 P. Kuson, A. Terdwongworakul / Journal of Food Engineering 116 (2013) 50–56

proximity of 41 kHz and dropped off as the frequency continued toincrease. The frequency of 41 kHz was thus subsequently definedas the peak frequency at which maximum reactance was observed.

At high frequencies (the left side of Fig. 5), both electrical valueswere similar as the fruits were more mature. However, at low fre-quencies, a decrease in both values was discerned with maturity,particularly in the case of CS.

Generally, the plot approximated a semicircular pattern for allstages of maturity. The arc shape was consistent over the measuredrange of maturity. As the fruit matured, the arc of the lower stemand the stem cross section continuously contracted which wassimilar to the ripening of nectarine fruit (Harker and Maindonald,1994) and the loss of moisture content in sliced cucumber (Liu,2006). Over the course of maturity, the contraction of the arc forthe stem cross section increased with the DAA value (Fig. 5b).However, contraction of the arc in the case of the lower stemwas not consistent with an increase in the DAA value (Fig. 5a) as

the circular arc of the 131 DAA was larger than the arc of the124 DAA. In the rind, the circular arc expanded as the fruit maturedwhich was contrary to the lower stem and the stem cross section(Fig. 5c). During growth, the EIS arc of fresh tea leaf was reportedto become greater (Mizukami et al., 2007). In addition, there wasa slight difference in the size of the arc of the rind among the matu-rity stages compared to those of the lower stem and the stem crosssection.

According to the above mentioned relationship, the differencein the size of the arc with maturity stages implies that the size ofthe arc could be used to represent maturity level. The arc size couldbe characterized by R values at a maximum (Rmx) frequency of200 kHz and minimum frequency (Rmn) at 1 kHz and X values atmaximum (Xmx) and minimum frequencies (Xmn) as well as atthe peak frequency of 41 kHz (Xpk) where the X value was maximal.The optimal frequencies selected were 1, 41 and 200 kHz,respectively.

Table 2Loading weight of first latent variable of partial leastsquares regression model for prediction of dry mattercontent of durian.

Variable Loading weight

ZmxLP �0.031XpkCS �0.715CmnRI �0.475CmxRI �0.336RpkRI 0.253XmxRI 0.290

Z, impedance; X, reactance; R, resistance; C, capaci-tance; LP, lower stem, CS, stem cross section; RI, rind;ZmxLP, impedance of the lower stem at maximum fre-quency (200 kHz); XpkCS, reactance of the stem crosssection at peak frequency (41 kHz); CmnRI, capacitanceof the rind at minimum frequency (1 kHz).

Table 3Correlation matrix between electrical impedance parameters that were selected usingstepwise procedure.

%DM ZmxLP XpkCS CmnRI CmxRI RpkRI XmxRI

%DM 1 �0.198 �0.647 �0.419 �0.274 0.218 0.241ZmxLP �0.198 1 0.161 �0.440 �0.461 0.530 0.471XpkCS �0.647 0.161 1 0.032 �0.024 0.097 0.154CmnRI �0.419 �0.440 0.032 1 0.857 �0.924 �0.866CmxRI �0.274 �0.461 �0.024 0.857 1 �0.951 �0.923RpkRI 0.218 0.530 0.097 �0.924 �0.951 1 0.958XmxRI 0.241 0.471 0.015 �0.866 �0.923 0.958 1

ZmxLP, impedance at maximum frequency (200 kHz) of the lower stem; XpkCS,reactance at peak frequency (41 kHz) of the stem cross section; CmnRI, capacitanceat minimum frequency (1 kHz) of the rind; CmxRI, capacitance at maximum fre-quency (200 kHz) of the rind; RpkRI, resistance at peak frequency (41 kHz) of therind; XmxRI, reactance at maximum frequency (200 kHz) of the rind.

Fig. 6. Scatter plot illustrating relationship of measured and predicted dry mattercontent.

Table 4Discriminant analysis results showing performance of classification into two maturity cla

Electrical impedance parameters Actual group Correctly

ZmxLP, XpkCS, CmnRI, CmxRI, RpkRI and XmxRI Immature durian 91.7Mature durian 83.3Total 86.7

ZmxLP, CmnRI, CmxRI, RpkRI and XmxRI Immature durian 83.3Mature durian 83.3Total 83.3

ZmxLP, impedance at maximum frequency (200 kHz) of the lower stem; XpkCS, reactance afrequency (1 kHz) of the rind; CmxRI, capacitance at maximum frequency (200 kHz) of thmaximum frequency (200 kHz) of the rind.

P. Kuson, A. Terdwongworakul / Journal of Food Engineering 116 (2013) 50–56 55

3.3. Quantitative evaluation of dry matter content

Regarding previous exploration of change in the impedance andcapacitance with the frequency as well as plot of the resistanceagainst reactance, the above mentioned electrical parameters (X,R, Z and C) at the maximum (mx), peak (pk) and minimum (mn)frequencies (200 kHz, 41 kHz and 1 kHz, respectively) of the lowerstem (LP), the cross section of the stem (CS) and the rind (RI) wereused as the independent variables of the PLSR models. Thus, RmnLPrepresented the resistance of the lower stem at the minimumfrequency.

The prediction performance of the model based on the above-mentioned electrical parameters was moderate with a correlationcoefficient of 0.798 and RMSEP of 4.63%. The model required threelatent variables to explain 69% variability of the dry matter contentbased on 73% variability of the EIS parameters. Although the accu-racy of the prediction was only fair, when considering that the pre-dictors were acquired from indirect measurement, theperformance could be deemed appropriate.

With the purpose of assisting the further development of apractical grading machine, simplification of the model by reducingthe number of predictors in the model using stepwise regression toselect the useful EIS parameters was studied.

The significant predictors selected as a result of the stepwisemethod were ZmxLP, XpkCS, CmnRI, CmxRI, RpkRI and XmxRI. These se-lected parameters were then submitted to PLSR analysis to modelthe maturity of the durian with respect to the dry matter content.The prediction of the dry matter content was improved by the sim-plified model, obtaining a correlation coefficient of 0.831 andRMSEP of 4.25%. However, five latent variables were required toexplain 72% of variance in the dry matter content. Regarding theloading weight of the first latent variable, which explained 38%variance in the dry matter (Table 2), XpkCS was the greatest con-tributing variable to the model which corresponded to the highestcorrelation coefficient with the dry matter content (r = �0.647 inTable 3). The second contributing variable was CmnRI. In a biologi-cal system, reactance is related mainly to capacitance which is ameasure of the biological membrane (Jackson and Harker, 2000).This implies that the difference in the dry matter content of thepulp during maturity stages was predominantly related to a varia-tion in the membrane of the rind and the stem cross section. Theassociated membranes were the plasma membrane and tonoplast(Bauchot et al., 2000). Fig. 6 illustrates the scatter plot of measureddry matter against predicted dry matter.

3.4. Maturity classification of durian

Further investigation was carried out to explore the applicationof the EIS parameters for maturity classification. This applicationaimed to separate the properly mature durians from the immaturedurians. Discriminant analysis was performed using the six EIS

sses based on six and five electrical impedance parameters.

classified durian (%) Predicted group

Immature durian Mature durian Total

11 1 123 15 18

14 16 3010 2 12

3 15 1813 17 30

t peak frequency (41 kHz) of the stem cross section; CmnRI, capacitance at minimume rind; RpkRI, resistance at peak frequency (41 kHz) of the rind; XmxRI, reactance at

56 P. Kuson, A. Terdwongworakul / Journal of Food Engineering 116 (2013) 50–56

parameters previously selected by the stepwise regression as theclassifying variables.

The classification model using the six EIS parameters performedfairly well, obtaining an accuracy of 86.7% (Table 4). However, theclassification model restricted the practical application to sortingafter harvest since measurement on the cross section of the XpkCSwas required. This meant durian had to be cut and the upper stemhad to be removed prior to measurement of the stem cross sectionfor the XpkCS, which made this technique unsuitable for monitoringmaturity of the fruit on the tree. Therefore, in order to enable theapplication of this technique to the harvest of the properly maturedurians on trees, the classification model was rebuilt using onlyfive EIS parameters excluding the XpkCS. After these adjustments,the classification performance dropped to 83.3% accuracy. How-ever, the five EIS parameters (ZmxLP, CmnRI, CmxRI, RpkRI, and XmxRI)could be measured on the rind and the stem while the fruit wasdeveloping on the trees.

4. Conclusions

Electrical impedance data of the stem and the rind of the durianproved to be suitable parameters to predict maturity with refer-ence to dry matter content. The parameters which were selectedby means of stepwise regression were shown to be optimal asthe model prediction accuracy improved using all parameters.The impedance of the stem cross section contributed the most tothe model and the capacitance of the rind complemented theprediction.

However, in the classification into immature and maturegroups, the classifying model based on five EIS parameters (ZmxLP,CmnRI, CmxRI, RpkRI, and XmxRI) yielded a fairly good performance.The overall accuracy was 83.3%. In addition, it was possible touse the five EIS parameters to measure the fruit developing onthe trees in a minimally destructive way. This facilitated mini-mally-destructive harvesting. Although accuracy of the predictionwas moderate, the technique was simple requiring measurementat only three frequencies on the stem and the rind. In addition,measurement on the rind and the stem provided an indirect mea-surement of the pulp. As capacitive reactance (X) is a function ofcapacitance (C), only one variable (X or C) needs to be measured.

Although two-terminal configuration was used in the frequencyrange of 1–200 kHz to minimize polarization impedance, there re-mained contact resistance which caused error in the measurement.The contact resistance can partially be reduced by choosing a suit-able metal electrodes or electrode surface of the material. In addi-tion, further investigation could be undertaken to improve theaccuracy of impedance measurement using a four-terminal config-uration (HIOKI four terminal probe 9140-10), which is an efficienttechnique to minimize electrode polarization impedance. In addi-tion, since the temperature of the samples during measurementswas an important factor, future investigation should take the tem-perature of individual samples into account for analysis.

Acknowledgements

The authors gratefully acknowledge the financial support fromthe Thailand Research Fund through the Royal Golden JubileePh.D. Program (Grant No. PHD/0118/2550) to student’s initials(Pramote Kuson) and advisor’s initials (Anupun Terdwongworakul).The Postharvest Technology Innovation Center (PHTIC), HigherEducation Commission, Center of Advanced Studies in IndustrialTechnology, Faculty of Engineering, and Center of Excellence onFood Agricultural Machinery (Kasetsart University) are alsoacknowledged for their supports of this research.

References

Ackmann, J.J., Seitz, M.A., 1984. Methods of complex impedance measurements inbiological tissues. CRC Critical Reviews Biomedical Engineering 11, 281–311.

Angersbach, A., Heinz, V., Knorr, D., 1999. Electrophysiological model of intact andprocessed plant tissues: cell disintegration criteria. Biotechnology Progress 15,753–762.

Bauchot, A.D., Harker, F.R., Arnold, W.M., 2000. The use of electrical impedancespectroscopy to assess the physiological condition of kiwifruit. PostharvestBiology and Technology 18, 9–18.

Beck, C.B., 2010. An Introduction to Plant Structure and Development: PlantAnatomy for the Twenty-First Century. Cambridge University Press, Cambridge,pp. 17–24.

Chattavongsin, R., Siriphanich, J., 1987. Fruit-stem anatomy of durians at variousharvesting stages (in Thai with English Abstract). In: Proceedings of the 26thAcademic Conference, Kasetsart University, Bangkok, February 3–5, 1987, pp.405–412.

Cunningham, T., 2000. Durian market report. Project ALA-97/68, UplandDevelopment Programme in Southern Mindanao, Philippines (UDP).

Ferris, C.D., 1974. Introduction to Bioelectrodes. Plenum Press, New York, pp. 235–239.

Gourshete, S.V., Chougule, C.A., Bagal, U.R., 2005. System development for bio-electrical impedance measurement for body composition analyzer. In: NationalConference on Electronics and Telecommunication, February 24–25, College ofEngineering, Osmanabad, India.

Harker, F.R., Dunlop, J., 1994. Electrical impedance studies of nectarines duringcoolstorage and fruit ripening. Postharvest Biology and Technology 4 (1–2),125–134.

Harker, F.R., Maindonald, J.H., 1994. Ripening of nectarine fruit. Plant Physiology106, 165–171.

Jackson, P.J., Harker, F.R., 2000. Apple bruise detection by electrical impedancemeasurement. HortScience 35 (1), 104–107.

Koksungnoen, O., Siriphanich, J., 2005. Anatomical changes during development ofdurian fruit cv. Kradum and Monthong. Thesis for Master of Science,Postharvest Technology, Interdisciplinary Graduate Program, KasetsartUniversity, Bangkok, Thailand.

Kongrattanaprasert, W., Arunrungrusmi, S., Pungsiri, B., Chamnongthai, K., Okuda,M., 2001. Nondestructive maturity determination of durian by force vibrationand ultrasonic. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems 9 (6), 703–719.

Liu, X., 2006. Electrical impedance spectroscopy applied in plant physiology studies.Thesis for Master of Engineering, School of Electrical and Computer EngineeringScience, Royal Melbourne Institute of Technology University, Australia.

Mizukami, Y., Yamada, K., Sawai, Y., Yamaguchi, Y., 2007. Measurement of fresh tealeaf growth using electrical impedance spectroscopy. Agricultural Journal 2 (1),134–139.

National Bureau of Agricultural Commodity and Food Standards, 2003. ThaiAgricultural Standard TAS 3-2003: Durian. Ministry of Agriculture andCooperatives, Thailand.

Neamsorn, N., Terdwongworakul, A., 2004. Maturity index of ‘Montong’ durianbased on stem strength and acoustic frequency response. In: Proceedings ofJapan–Thailand Joint Symposium on Nondestructive Evaluation Technology,May 18–21, Bangkok, Thailand.

Peres, R.S., Cassel, E., Azambuja1, D.S., 2012. Black wattle tannin as steel corrosioninhibitor. Laboratorio de Eletroquımica, Instituto de Quımica, UFRGS, AvenidaBento Gonc�alves 9500, 90619-900 Porto Alegre, RS, Brazil. InternationalScholarly Research Network ISRN Corrosion Volume 2012, ID 937920, p. 9.

Rehman, M., Izneid, B.A.J.A., Abdullah, M.Z., Arshad, M.R., 2011. Assessment ofquality of fruits using impedance spectroscopy. International Journal of FoodScience and Technology 46, 1303–1309.

Rutpralom, T., Kumhom, P., Chamnongthai, K., 2002. Nondestructive maturitydetermination of durian by using microwave moisture sensing. In: IEEEInternational Conference on Industrial and Technology (ICIT), Bangkok,Thailand, December 11–14, 2002, pp. 155–158.

Sangwanangkul, P., Siriphanich, J., 2000. Growth and maturation of durian fruit cv.Monthong. Thai Journal of Agricultural Science 33 (1–2), 75–82.

Siriphanich, J., Jerapat, S., 2005. Use of dry matter as maturity index in Kradumdurians (Durio Zibethinus Murr.). In: Information and Technology forSustainable Fruit and Vegetable Production FRUTIC 05, September 12–16,2005, Montpellier France.

Siriphanich, J., 2011. Durian (Durio zibethinus Merr.). In: Yahia, E.M. (Ed.),Postharvest Biology and Technology of Tropical and Subtropical Fruits.Volume 3: Cocona to Mango. Woodhead Publishing, UK.

Stout, D.G., 1988. Effect of cold acclimation on bulk tissue electrical impedance 1.Measurements with birdsfoot trefoil at subfreezing temperatures. PlantPhysiology 86, 275–282.

Sugiyama, J., Hayashi, T., Horiuchi, H., 1987. Electrical impedance of kiwifruit.Nippon Shokuhin Kogyo Gakkaishi 33, 725–730.

Zhang, M.I.N., Stout, D.G., Willison, J.H.M., 1990. Electrical impedance analysis inplant tissues: symplasmic resistance and membrane capacitance in the Haydenmodel. Journal of Experimental Botany 41, 371–380.