Pasting properties by near-infrared re ° ectance analysis of whole grain paddy rice samples

Pornarree Siriphollakul*‡, Sirichai Kanlayanarat*, Ronnarit Rittiron†||, Jaitip Wanitchang, Thongchai Suwonsichon, Panida Boonyaritthongchai* and Kazuhiro Nakano¶ *Division of Postharvest Technology KingMongkut's University of Technology Thonburi 126 Pracha-Utid Road Bangmod, Toongkru, Bangkok 10140, Thailand Department of Food Engineering, Faculty of Engineering at Kamphaengsaen Kasetsart University, Nakhonpathom, Thailand Division of Agricultural Engineering and Technology Rajamungala University of Technology Tawan-ok, 43/6 Bangpha, Sriracha, Chonburi, Thailand Department of Product Development, Kasetsart University 50 Ngam Wong Wan Rd Ladyao, Chatuchak Bangkok, Thailand ¶Graduate School of Science and Technology, Niigata University 8050 Ikarashi 2-no-cho Nishi-ku, Niigata, 950-2181 Japan ||fengror@ku.ac.th


Introduction
Typically, the most abundant component in rice is starch.It constitutes more than 80% of the total dry weight of polished rice.Starch is an important factor for determining the quality of rice products. 1n many Asian foods, such as noodles rice starch is the crucial ingredient.Starch pasting properties in°uence noodle quality, and are also directly responsible for starch industrial uses. 2 Conventionally, the pasting properties of starch are measured by the rapid visco-analyzer (RVA).The analytical instrument was initially used to determine preharvest damage in cereals.The RVA had been used to provide information about pasting characteristics of starch of a particular sample under analysis (e.g., pasting peak viscosity (PV), ¯nal viscosity). 3One of the most important indices that are used to explain rice aging are pasting properties.During storage, changes of pasting properties are correlated to aging process individually. 4In terms of rice storage, changing trends of rice paste viscosity are di®erent. 5or instance, the viscosity of rice paste increases considerably after short-to intermediate-term storage (months) of milled rice. 6However, it decreases during long-term storage (years). 7For RVA technique, in order to measure the pasting properties of rice, several steps of sample preparations and instruments are of major concern, since the method is time consuming and laborious.
A rapid technique, near-infrared re°ectance spectroscopy (NIRS), was developed for measuring the quality characteristics that are routinely tested in cereals. 8In comparison to routine, conventional analysis methods, NIRS has many advantages.Firstly, sample preparation is simple and quick.Moreover complicated treatments or techniques are not needed for measurements.Second, it is a realtime determination of multiple characters with the development of calibration models.Thirdly, it is more °exible since the method can be applied to measure several kinds of samples.Fourthly, materials can be shipped and reused after measurements because the method is nondestructible.Lastly, the technique is economical; reagents and/or manpower are not needed.It is a clean and environmentfriendly analytical system since no wastes emerge. 9 few reports have been published on establishing the NIRS model for evaluating rice characteristics.Delwiche et al 10 successfully used NIRS to determine milled whole-grain samples.Likewise, Bao et al. 11 also used visible re°ectance and NIRS for determining milled rice °our samples and achieved satisfying prediction results.Moreover, Natsuga and Kawamura 12 studied pasting properties of brown and milled rice samples by visible re°ectance and NIRS, and reported that the technique was an accurate tool for predicting pasting properties.Based on the aforementioned works, no previous research has reported the application of NIRS in paddy rice.
Therefore, this study was focused on Khoa Dawk Mali 105 paddy rice (cultivar KDML 105) as a material.The Khao Dawk Mali 105 rice is the most common rice cultivar in Thailand.It has a unique aroma and mild taste and is internationally recognized as \Jasmine Rice".In Thailand, KDML 105 rice is considered to be a vital crop for domestic consumption and primary export commodity for agricultural economic sector. 13Thailand is one of the world's largest KDML 105 rice producers and also the world's largest rice exporter in terms of rice amounts and cultivation area.Normally, the rice mills purchase paddy rice from farmers in order to process it into brown, parboiled and milled rice for export and domestic consumption according to market demands.Before trading and purchasing, qualities must be monitored for paddy pricing.Therefore, in this study, NIRS was used to measure pasting properties of paddy rice form, focusing on a nondestructive technique for measuring paddy rice quality.

Rice samples
264 rice samples (Khao Dawk Mali (KDML105) variety) were harvested from the Rice Seed Center in Chonburi Province which is the state agency of Thailand.Then, the samples were dried under the sun for a day.Moisture content of paddy rice samples ranged from 11% to 14% on wet basis.Subsequently, the samples were stored in a silo which had dimensions of 1:5 Â 1:5 Â 4:5 m (width Â length Â height), without controlling temperature.All samples (1000 kg) were kept throughout the storage period (6 months) after harvesting to study the change of pasting properties after the storage and extend ranges of pasting properties.Samples were taken once a month for the ¯rst 2 months of storage period.Afterward they were taken every 15 days until the end of the storage.

Spectroscopy analyses
Entire paddy rice samples were scanned on a near infrared spectrophotometer (The InfraScan, BRUINS INSTRUMENTS, Puchheim, Germany).Approximately 70 g of a sample were loaded in a small ring quartz window-clad cylindrical cell (diameter ¼ 105 mm, height ¼ 15 mm).The re°ectance was measured in the NIR region within 1400-2400 nm and recorded during the rotation of the cylindrical cell at 0.5 nm intervals.Each sample was subsequently scanned, and average spectrum was collected (Fig. 1).

Pasting properties
After spectral acquisition, paddy rice samples were de-husked, milled, ground and sifted through a 0.8 mm sieve screen.Rice °our sample ($ 4.0 g of °our corrected using the moisture content of the sample, AE 0.01 g) was mixed with distilled water ($ 25.0 g as the function of the amount of adjusted sample AE 0.01 g) in an aluminum cylinder.The mixture was agitated by raising and lowering the plastic paddle through the aluminum cylinder before inserting the cylinder into the instrument.The pasting properties of slurries were determined with a rapid visco analyzer (RVA-4 Model, Newport Scienti¯c Pty., Ltd., Warriewood NSW, Australia) in the RVU unit by performing the viscosity pro¯le during 12.5-min heating cycle.The test pro¯le had a starting temperature of 50 C, which was maintained for 1 min, later raised to 90 C within 4 min, maintained for 10 min, then cooled to 50 C in 1 min, and maintained for 1 min, with a stirring speed of 160 rpm as in the test period.Properties were recorded and used for calculating other pasting parameters such as PV, hot paste viscosity (HV), cold paste viscosity (CV), breakdown (BD), setback (SB) and consistency (CS). 14

Calibration and validation procedures
Before calibration development, the reference value outliers were identi¯ed by Z-score.The outliers were limited to below 5% of samples unless the excess of outliers was removed.Spectral preprocessing was performed and calibration was developed based on partial least squares (PLS) regression with test set of a sample, using the software for multivariate analysis (Unscrambler version 9.8).A total of 160 samples were used as a calibration set and the rest (104 samples) was used as a validation set.Statistic terms such as correlation of coe±cients (RÞ; ratio of standard error of prediction (SEP) to standard deviation (RPD); SEP and standard deviation (SD) of di®erence between measured and predicted values; bias or average of di®erence between measured and predicted values of validation set were calculated.Moreover, validation test set was checked for reliability by following guidelines in ISO 12099:2010 15 as follows: Firstly, t-test is checked by the signi¯cance of the bias.The calculation of the bias con¯dence limits (BCLS), the limit for accepting or rejecting model, is determined by T b which was calculated according to Eq. (1).If bias was smaller than T b , then the bias obtained from the model had no signi¯cance or NIRS predicted values were not signi¯cantly di®erent form measured values.
where is the probability of making a type-l error; tis the appropriate Student t-value for a twotailed test with degrees of freedom associated with SEP and the selected probability of a type-I error; nis the number of independent samples; SEPis the standard error of prediction.
Secondly, the validity of the calibration model was checked by F -Test (ratio of variances) for SEP.The unexplained error con¯dence limits (UECLS), T UE were calculated according to Eq. ( 2).If SEP was smaller than T UE obtained from the SEP was in an acceptable range.
where SECis the standard error of calibration; ¼ n v À 1is the numerator; degrees of freedom associated with SEP of the test set in which n is the number of samples in the validation process; M ¼ n c À p À 1is the denominator; degrees of freedom associated with SEC n cis the number of calibration samples, pis the number of terms or PLS factors in the model.
Lastly, the slope, b of the sample regression: y ¼ a þ bŷ with the reference value as the dependent variable and the predicted NIRS values as the independent variable was checked.From the least squares ¯tting, the slope is calculated as: where S ŷyis the covariance between measured and predicted values; S 2 ŷis the variance of the n predicted values.The intercept is calculated as: where ŷis the mean of the predicted values, yis the mean of the measured values, bis the slope.
As for the bias, a t-test can be calculated to check the hypothesis that b ¼ 1 where S 2 ŷis the variance of the n predicted values S resis the residual SD as de¯ned in Eq. ( 6) in which ais the interception from Eq. ( 4), bis the slope from Eq. ( 3), y iis the ith measured value, ŷiis the ith predicted value obtained when applying the multivariate NIRS model.
The slope, b, is considered as signi¯cantly di®erent from 1 when t obs !t ð1À=2Þ where t obsis the observed t-value, calculated according to Eq. ( 5), t ð1À=2Þis the t-value obtained for a probability of ¼ 0:05 (5%).

Pasting properties change
Pasting properties and sensitive index of the intrinsic properties of starchy rice materials are important.They ideally provide an excellent re°ection of any changing phenomena that occur in the grain during storage.Changes of pasting properties are shown in Fig. 2. The PV, one of pasting properties, continually rose during the ¯rst 140 days and gradually declined until the end of studied period.For instance, in freshly harvested rice, the increase in PV during early period of storage (few months) was in°uenced by the presence of amylase activity.However, later the PV gradually declined due to the lack of amylase activity. 16In case of the SB, it notably increased during 185 days of storage and slightly changed afterward.A similar trend was observed for the cold paste viscosity, after storing for 185 days, the value declined and was close to the value observed in fresh rice.There was a signi¯cant change in BD at all storage times.This change can be attributed to the uniqueness of the starch granules.The decrease in BD value indicated that the ability of the starch granules to BD after cooking signi¯cantly reduced by aging of the granules.The increase in PV showed that the starch granules of storage rice were more resistant to swelling than those of fresh rice.
Table 1 shows the mean, minimum, maximum and SD values of the pasting properties of rice samples.Typically, in order to develop an e®ective model for quality prediction, the variations such as cultivars and harvest time should included in datasets.In this experiment, we used only one cultivar, but the data variations were dependent on storage time.

Pretreatment procedure
The original NIR re°ectance spectra of paddy rice samples in long wavelength region (1400-2400 nm) are shown in Fig. 3.According to the chemical assignments, H 2 O molecule could absorb at 1450 nm (O-H stretching 1st overtone) and 1940 nm (O-H stretching þ O-H deformation).As for starch molecule, the chemical bond vibrations were also observed at 1450 and 2100 nm (2 Â O-H deformation þ 2 Â C-O stretching) previously described by Osborne et al.In our study, spectral characteristic notably revealed baseline shift phenomenon.In addition, after the radiation encountered discrete particle within the sample, destructive interference became incomplete and the radiation propagated in all directions.It is known as scattering.This was due to the random di®usion in re°ectance measurements since the radiation is re°ected, di®used and scattered at further sample interface.In case of rice samples, macrostructure of rice intensively a®ected the random optical path length because the light was transmitted to the rice and re°ected randomly among rice kernels which led to scattering.Therefore, various NIR spectra pretreatments were employed in this study, namely Savitzky-Goley ¯rst derivative and standard normal variant (SNV).

PLS results
The results of PLS models for predicting pasting properties from paddy rice spectra are shown in Table 2. Performance was evaluated using the correlation coe±cient (R) between measured values and predicted NIRS values; the SEP; RPD and bias.The best PLS of a model developed for predicting SB was obtained from the Savitzky-Golay smoothing (gap 1.25 nm) and SNV in 1400-2250 nm range.A value of R of pasting properties of rice (SB) gave good results, providing high R (0.96), small SEP (6.54) and low bias (0.58).Among them, higher RPDs (> 2) were mostly obtained for PV (2.34), CV (2.02), BD (2.14), SB (3.52) and CS (2.30).Only HV (1.82) had a relatively low RPD.The RPD was explained by Fearn. 17Calibration of calculated RPD between 2 and 10 proved to be a good model resulting in dependable predicting result.The results in Table 2 also indicated those reasonable models (R > 0:8) that were obtained from the use of wavelength ranges regions between 1400 and 2250 nm.This region typically has high  molar absorptivity.The regression coe±cients of the developed model for predicting SB using six factors are presented in Fig. 4. The regression analysis also revealed the highest relationship at 1527 and 2099 nm wavelength.Osborne 8 reported that wavelengths of 1528 and 2100 nm are typically responsible for starch, the major component of rice.The pasting quality of rice is basically dictated by the quality of the starch.Pasting properties e®ectively explained the behavior of rice °our and starch during processing (heating and/or cooling).For other studied properties, absorbance of starch was shown to be highly in°uenced in regression coe±cient plots of all models.Figures 5(a  t obs < tð1 À =2Þ showed a successful prediction or b was looked upon as 1.Our results showed that slopes of all models were not signi¯cantly di®erent from 1. Therefore, developed models for predicting pasting properties of paddy rice were accomplished according to ISO 12099.

Conclusion
Nondestructive measurement of pasting properties of paddy rice was studied using NIRS.The technique performed well for predicting the pasting properties with satisfactory performance as the validity of the calibration models was statistically tested.This study proved that the use of NIRS is suitable for predicting paddy rice pasting properties.

Fig. 1 . 3 J
Fig. 1.Schematic diagram of re°ectance NIRS measurement displaying in top view and side view.

Fig. 2 . 4 J
Fig. 2. The change of pasting properties in KDML105 cultivar after storage.

Fig. 4 .Fig. 5 .
Fig. 4. The PLS model coe±cients of model developed for predicting setback of paddy rice in wavelength range of 1400-2400 nm.

Table 1 .
Pasting properties of rice samples throughout six months storage period.

Table 2 .
PLS models result for predicting pasting properties of paddy rice using di®erence mathematical pretreatments and wavelength range selections.Note: R; correlation of coe±cient, SEP; standard error of prediction, RPD; ratio of standard error of prediction to standard deviation and bias; average error of prediction.

Table 5 .
The statistic test of slope of models development.Slope obtained from the regression of the measured values and the predicted values.The t-value obtained for a probability of ¼ 0:05 (5%) when t obs < tð1 À =2Þ, slope is not signi¯cance from 1.
a b The observed t-value.c