• About Us +
• For Contributors +
• Journal Search +
Journal Search Engine
ISSN : 1598-5504(Print)
ISSN : 2383-8272(Online)
Journal of Agriculture & Life Science Vol.53 No.3 pp.147-157
DOI : https://doi.org/10.14397/jals.2019.53.3.147

# Feasibility Study for an Optical Sensing System for Hardy Kiwi (Actinidia arguta) Sugar Content Estimation

Sangyoon Lee1, Shagor Sarkar1, Youngki Park2, Jaekyeong Yang3, Giyoung Kweon1*
1Department of Bio-Industrial Machinery Engineering, Gyeongsang National University (Institute of Agriculture and Life Science), Jinju, 52828, Korea
2Forest Medicinal Resources Research Center, National Institute of Forest Science, Yeongju, 36040, Korea
3Department of Environmental Materials Science, Gyeongsang National University (Institute of Agriculture and Life Science), Jinju, 52828, Korea
Corresponding author: Giyoung Kweon Tel: +82-55-772-1898 Fax: +82-55-772-1899 E-mail: gkweon@gnu.ac.kr
March 11, 2019 April 2, 2019 April 9, 2019

## Abstract

In this study, we tried to find out the most appropriate pre-processing method and to verify the feasibility of developing a low-price sensing system for predicting the hardy kiwis sugar content based on VNIRS and subsequent spectral analysis. A total of 495 hardy kiwi samples were collected from three farms in Muju, Jeollabukdo, South Korea. The samples were scanned with a spectrophotometer in the range of 730-2300 nm with 1 nm spectral sampling interval. The measured data were arbitrarily separated into calibration and validation data for sugar content prediction. Partial least squares (PLS) regression was performed using various combinations of pre-processing methods. When the latent variable (LV) was 8 with the pre-processing combination of standard normal variate (SNV) and orthogonal signal correction (OSC), the highest R2 values of calibration and validation were 0.78 and 0.84, respectively. The possibility of predicting the sugar content of hardy kiwi was also examined at spectral sampling intervals of 6 and 10 nm in the narrower spectral range from 730 nm to 1200 nm for a low-price optical sensing system. The prediction performance had promising results with R2 values of 0.84 and 0.80 for 6 and 10 nm, respectively. Future studies will aim to develop a low-price optical sensing system with a combination of optical components such as photodiodes, light-emitting diodes (LEDs) and/or lamps, and to locate a more reliable prediction model by including meteorological data, soil data, and different varieties of hardy kiwi plants.

## 초록

Korea Forest Service
FTIS 2017091B10-1919-AB01

## Introduction

With the recent increases in national income and well-being in Korea, Korean consumers are giving more importance to the safety and quality of food rather than the quantity or price (Huh & Park, 2007). As such, farming policies focused on increasing food production are now focused on enhancing the understanding of the interests of domestic consumers. Moreover, it has become more desirable to expand the income opportunities of farmers by creating new markets through high-quality agroforestry products to satisfy consumer requirements (Kunznets, 1955;Aukrust, 1970).

In general, the elements that determine food quality include the appearance, texture, flavour, nutritional value, and food safety (Wolpert et al., 1983). The sugar content, a major factor affecting the flavour, is also an important element in determining the quality of fruit, while also being one of the major criteria based on which consumers select fruits (Gautier et al., 2008;Harel-Beja et al., 2010).

The hardy kiwi (Actinidia arguta), one of the major honey fruit plants in Korea, is a deciduous broad-leaf climbing vine native to Korea, China, and Japan that belongs to the Actinidiaceae family and grows in the valleys of mountainous regions (Kim et al., 2018). Hardy kiwis have a strong resistance to both cold weather and damage from disease and insects, which allows them to be grown in all regions of Korea. They also have a faster maturation period than regular kiwis (Sohn et al., 2017). Following development and safety assessments, the Korean National Institute of Forest Science is now supplying new varieties of hardy kiwis, such as the Saehan, Daesung, Chilbo, and Autumn Sense varieties. Hardy kiwi fruits are small, rich in taste, aromatic, and edible. In addition, the hardy kiwi is known to possess excellent medicinal effects, having been demonstrated to reduce fever, quench thirst, and alleviate indigestion (Sohn et al., 2017). Furthermore, due to its high vitamin C content, it is known to have excellent antioxidant properties (An et al., 2016) in addition to a therapeutic effect on scurvy patients (Ha et al., 2015). As such, the supply of and demand for hardy kiwis are increasing, and as the scale of their cultivation grows, systematic cultivation management technologies are required (Bieniek et al., 2016). In particular, the current situation requires analysis of the optimal growth conditions of hardy kiwis that are known to affect the product quality, including the sugar content.

Currently reported methods for estimating sugar contents can be divided into two main categories. The conventional method for measuring sugar content involves the breakage of a few arbitrarily sampled kiwi fruits and chemical test of samples. However, this method is slow and the fruits used for testing are destroyed and cannot be sold later (Bureau et al., 2009). Numerous studies have therefore focused on non-destructive methods for the rapid measurement of fruit sugar contents (Kalaj et al., 2018). Such methods include nuclear magnetic resonance (Ohno et al., 1984), X-ray imaging (Stevens & Schroeder, 2009), tomography (Kawachi et al., 2011), and near-infrared spectroscopy (NIRS, Rodriguez-Saona et al., 2001). Among these methods, NIRS is currently the most commonly used method in fruit-selection systems (Rambla et al., 1997;Hu et al., 2017).

This method allows qualitative and quantitative analysis of the fruit quality within 1 min in a nondestructive manner, and large quantities of samples can be accurately analyzed within a short period of time, this renders an essential analytical method to aid in fruit harvesting (Pissard et al., 2018). Visible or near-infrared instruments produce a large amount of spectral data that contain important analytical information (Golic & Walsh, 2006). The spectral data need to be pre-processed to reduce the effect of irrelevant information such as background and noise spectra. The pre-processing also helps to obtain accurate, reliable and stable calibration models from the raw data (Blanco & Villarroya, 2002). To improve the multivariate linearity of absorbance data several pre-processing methods can be used such as standard normal variate (SNV), orthogonal signal correction (OSC), multiplicative scattering correction (MSC), first, and second derivative. However, the application of the NIRS method for commercial use is limited, as the required high-resolution spectrophotometers are quite expensive (Jenny et al., 2002). Thus, studies need to implement for development of low-price instrument with the combination of optical components such as light-emitting diodes (LEDs), photodiodes, lasers, and lamps and lenses with visible (VIS) and near-infrared wavelengths.

The objectives of this study were to investigate the most suitable pre-processing for hardy kiwi sugar content prediction model and to investigate the possibility of using a low-price optical sensing system by examining various wavelength ranges and spectral sampling intervals.

## Materials and Methods

### 1 Hardy kiwi sample preparation

A total of 495 hardy kiwi samples were collected from three different farms within Muju County, Jeollabukdo, South Korea-165 each from the Mujueup (Farm 1), Seolcheon-myeon (Farm 2), and Jeoksang- myeon (Farm 3) areas. Figure 1 shows hardy kiwis and a hardy kiwi orchard. The hardy kiwis sampled for this study were the Autumn Sense and Daesung varieties. Subsequently, the collected hardy kiwi samples were measured using a spectrophotometer, where a total of 495 measured data sets were arbitrarily divided into two groups for calibration and validation (prediction).

### 2 Sugar content measurements

Sugar contents are expressed in degree Brix, which represents 1 g of sucrose dissolved in 100 g of water, and so the concentration of sugar per 100 g of water is referred to as the sugar content. Generally, two types of refractometers (i.e., optical and digital refractometers) are used to measure sugar content. Refractometers often are of the type based on measurement of the so-called critical angle of total reflection, in which the position of a boundary or shadow line dividing a field of view into a bright and a dark portion is observed through an eyepiece against a fixed scale or mask. It would be highly desirable to provide an automatic refractometer capable of completely automatic measurements of refractive indexes to eliminate the possibility of human error arising from manual alignment through an eyepiece (Michalik & Sloan, 1987). In this study, a portable digital refractometer (SCM-1000, HM Digital, Korea) with a measurement range of 0-55% Brix was used to measure the sugar contents of the hardy kiwi samples.

The sample calibration data indicated a mean sugar content of 16.0, a standard deviation (SD) of 3.92, a maximum value of 25.1, and a minimum value of 10.1 Brix. The validation data indicated a mean sugar content of 16.1, an SD of 3.46, a maximum value of 23.4, and a minimum value of 10.3 Brix.

### 3 Visible and near-infrared spectroscopic analysis

The near-infrared region is located between the visible and infrared regions and covered the range of 780-3000 nm. The principle behind NIRS is the application of external light energy to an object and subsequent measurement of the altered energy radiated from the object for its physical, chemical, quantitative, and qualitative analyses. This method does not tend to require the use of reagents or processes such as quantification, mixing, heating, or extraction, and so it allows multiple components to be analyzed rapidly and simultaneously.

A FieldSpec 4 spectrophotometer (ASD Inc., Colorado, USA) was used in this study (Fig. 2). This high-resolution device spectral range covers wavelengths from 350 to 2,500 nm and the sampling rate is 0.1 seconds per spectrum. Spectral resolution ranges from 3 nm in the low wavelengths to 8 nm in the high wavelengths, this device records spectra based on information from 2,151 bands, and spectral sampling interval is 1.4 nm at 350-1000 nm and 1.1 nm at 1001-2500 nm. However, the spectral data for the hardy kiwi samples were used in the range of 730-2300 nm to locate optimal pre-processing following exclusion of the wavelength ranges with low correlations to the predicted sugar content and noise in high range. Other analyses were also performed in a narrower range up to 1200 nm with spectral sampling intervals of 1, 6, and 10 nm to investigate the possibility of analysis using a low-price optical sensing system, since spectrometers or optical components with high spectral resolution in the NIR area are quite expensive in general.

The acquired data were converted the absorbance data for the sugar content analyses. The absorbance was calculated as follows:

$a b s = l o g 10 ( r e f − d a r k s a m p l e − d a r k )$
(1)

Where abs represents the absorbance; ref is the reference light intensity, i.e., the light intensity of the light source employed; dark is the dark intensity, i.e., the light intensity recorded by the spectrophotometer with the light source switched off or blocked and without sample; and sample is the light intensity after passing through the sample, i.e., the light intensity recorded by the spectrophotometer when the light from the light source reflected from the sample of interest.

### 4 Sugar content prediction model

Various pre-processing methods are helpful in effective analyzing of spectral data. Among them, first derivative is widely used for analysis by various regression method which can remove drifting, scattering, background interference, distinguish superpose peaks, and strengthen the spectral resolution and sensitivity. OSC pre-processing also eliminates unnecessary signals before calibration by some orthogonal ways that effectively reduces the principal component (PC) number and strengthen the prediction ability of the calibration model. MSC pre-processing can remove the ideal linear scattering and effects well when the linear relationship between absorbance and sample concentration is good. This process works by regressing a measured spectrum against a reference spectrum and then corrected the measured spectrum by using the slope of this fit. Whereas, SNV preprocessing removes the deviations caused by particle size and scattering. The correction ability of SNV is stronger than the MSC (Blanco & Villarroya, 2002).

The obtained spectrum represents combined effects of internal quality of fruits. Therefore, the relationship between the components determined by NIRS and chemical analyses must be identified. Such a relationship can be determined using a calibration equation, which is derived using multivariate statistical analysis.

The prediction standard error of the model, the root mean square error of calibration (RMSEC, see Equation (2)), the standard error of validation of the unknown sample, the bias (see Equation (3)), the root mean square error of prediction (RMSEP, see Equation (4)), and the ratio of prediction to deviation (RPD) were used to test the predictability of the developed model. According to articles (Chang et al., 2001) and (Kweon, 2012), the RPD calculation indicates the precision of the prediction model, and the performance of the prediction model for NIR analysis can be classified as high, average, or low based on SD/RMSEP values of >2.0, 1.4-2.0, and <1.4, respectively. A higher RPD value indicates a more accurate prediction.

$R M S E C = ∑ i = 1 m ( y i − y ^ i ) 2 m − f − 1$
(2)

$b i a s = ∑ i = 1 m ∑ ( y i − y ^ i ) m$
(3)

$R M S E P = ∑ i = 1 m ( ( y i − y ^ i ) − b i a s ) 2 m − 1$
(4)

Where represents the sugar content of the sample, represents the predicted value based on the model, represents the degree of freedom (latent variable), and represents the number of samples.

The general criteria applied for comparing the performances of prediction regression models developed using different pre-processing methods and regression models are as follows (Noh, 2000): 1) The RMSEC and RMSEP should be small and the difference between these should also be small, 2) the prediction model should have a small bias, 3) the R2 values for the developed calibration and validation models should be large, and 4) The PLS latent variable number included in the model should be small.

In this study, MATLAB (MathWorks, USA) and PLS Toolbox (Eigenvector Research Inc., USA) was used to analyze the spectral data and to develop a sugar content prediction model. PLS regression was performed on the obtained spectral data after performing various combinations of above-mentioned preprocessing steps.

## Results and Discussion

### 1 Optimal pre-processing for hardy kiwi sugar content prediction

As shown in table 1, when the MSC+OSC or SNV+ OSC pre-processing combinations were applied, a good correlation was obtained when the latent variable (LV) was 8. Based on these results, it was determined that pre-processing involving MSC or SNV is required to compensate for the scattering effect, which is based on the physical properties of the fruit, while pre-processing with only OSC showed high prediction performance.

Of these two pre-processing methods, the SNV+ OSC (see Fig. 3) combination was considered the most suitable, as it met the selection criteria for the optimal prediction model explained in the Materials and Methods section. Here, the calibration R2, prediction R2, bias, and RPD values were 0.78, 0.84, -0.20, and 2.50, respectively, which indicate effective prediction performance.

### 2 The effect of spectral range and sampling interval on hardy kiwi sugar content prediction

Interestingly, PLS analysis of the spectral data obtained between 730 and 1200 nm showed rather superior performance to those obtained at wavelengths up to 2300 nm in the spectral sampling interval of 1 nm. We found that the best prediction was also obtained by the SNV+OSC pre-processing combination (see Table 2). The results exhibited that the calibration R2, prediction R2, bias, and RPD values were 0.90, 0.86, -0.15 and 2.63, respectively, which demonstrate excellent predictability.

As mentioned above, among various spectrophotometers with narrow wavelength ranges, cheaper instruments have significantly poorer resolutions. More specifically, high-specification spectrophotometers have a spectral sampling interval of ~1 nm, whereas lowprice ones generally have sampling intervals of 5-6 and 10 nm. Thus, PLS analysis was performed after pre-processing with a spectral sampling interval of 6 nm using an original spectral sampling interval of 1 nm dataset and selecting evenly spaced spectra (i.e., by 6 nm) for the data measured between 730 and 1200 nm. As shown in Table 3, the results exhibited a slightly different tendency, as spectral sampling interval of 6 nm, the best results were achieved using the MSC+OSC pre-processing combination, while the SNV+OSC combination, which previously gave the best results, resulted in the lowest LV value and relatively high prediction performance.

Similarly, Table 4 shows the results of PLS analysis with a spectral sampling interval of 10 nm. In this case, pre-processing with the SNV+OSC combination gave a relatively good result. Accordingly, the combination of SNV and OSC would be selected for pre-processing in future study of the sugar content prediction of hardy kiwis. From the data presented in tables 1, 2, 3 and 4, although the analysis at higher resolutions exhibited superior prediction performance, the results obtained at sampling intervals of 6 and 10 nm remained promising.

From the data presented in Fig. 4, it was apparent that the spectra with a lower resolution exhibited slightly poorer prediction performance and a lower slope value (sensitivity) according to the trend line equations. However, the R2 values for the results obtained with resolutions of 6 and 10 nm were 0.84 and 0.80 in a spectral range of 730-1200 nm. This was an encouraging result considering that wide sampling intervals and narrow range wavelengths were employed. Based on these results, it is feasible to predict the sugar content of hardy kiwis using a relatively low-price and low-specification sensing system, rather than expensive, high-specification instrument with a range beyond 1200 nm.

The prediction of the fruits sugar content is important to measure the fruit quality, as the sugar content affects the taste of the fruit and the overall fruit quality. In this study, the most appropriate preprocessing method was found out and the feasibility of developing a low-price sensing system was verified to predict the hardy kiwis sugar content based on VNIRS and subsequent spectral analysis.

Data were collected over a spectral range of 730-2300 nm with a spectral sampling interval of 1 nm, and partial least squares analysis was performed on the obtained data using a combination of standard normal variate and orthogonal signal correction preprocessing to predict the sugar content of hardy kiwis. The results showed high correlations between the calibration and prediction models with R2 values of 0.78 and 0.84 being calculated, respectively, while the ratio of prediction to deviation value, which indicates the accuracy of the prediction model, suggested an excellent predictability with a value of 2.50. Moreover, when PLS analysis was performed on the 730-1200 nm spectra with spectral sampling intervals of 6 and 10 nm, the prediction performance was promising as R2 of 0.84 and 0.80. Thereby the development of a low-price optical sensing system could be used to estimate kiwi sugar content.

## Acknowledgement

This study was carried out with the support of ‘R&D Program for Forest Science Technology (Project No. FTIS 2017091B10-1919-AB01)’ provided by Korea Forest Service (Korea Forestry Promotion Institute).

## Figure

Hardy kiwis (left) and a hardy kiwi orchard (right).

The spectrophotometer (left) and the spectral acquisition process for the hardy kiwi samples (right).

Relationship between the measured and predicted sugar contents for hardy kiwis for validation datasets in the spectral range of 730-2300 nm with spectral sampling interval of 1 nm.

Relationship between the measured and predicted sugar content at 730-1200 nm of spectral range with spectral sampling intervals of (a) 1 nm, (b) 6 nm, and (c) 10 nm.

## Table

Results of PLS regression analysis for the sugar content prediction of hardy kiwis based on the use of various pre-processing combinations (spectral range=730-2300 nm, spectral sampling interval=1 nm)

Results of PLS regression analysis for the sugar content prediction of hardy kiwis for the various preprocessing combination with a spectral range of 730-1200 nm and a spectral sampling interval of 1 nm

Results of PLS regression analysis for the sugar content prediction of hardy kiwis for the various preprocessing combinations with a spectral range of 730-1200 nm and a spectral sampling interval of 6 nm

Results of PLS regression analysis for the sugar content prediction of hardy kiwis for the various preprocessing combinations with a spectral range of 730-1200 nm and a spectral sampling interval of 10 nm

## Reference

1. An X , Lee SG , Kang H , Heo HJ , Cho YS and Kim DO. 2016. Antioxidant and anti-inflammatory effects of various cultivars of kiwi berry (Actinidia arguta) on lipopolysaccharide-stimulated RAW 264.7 cells. J. of Microbiol. and Biotechnol. 26: 1367-1374.
2. Aukrust O. 1970. PRIM I: A model of the price and income distribution mechanism of an open economy. The Rev. of Income and wealth 16: 51-78.
3. Bieniek A , Draganska E and Pranckietis V. 2016. Assesment of climatic conditions for Actinidia arguta cultivation in north-eastern Poland. Zemdirbyste- Agric. 103: 311-318.
4. Blanco M and Villarroya I. 2002. NIR spectroscopy: a rapid response analytical tool. Trend in Analytical Chem. 21: 240-250.
5. Bureau S , Ruiz D , Reich M , Gouble B , Bertrand D , Audergon JM and Renard CM. 2009. Rapid and nondestructive analysis of apricot fruit quality using FTnear- infrared spectroscopy. Food Chem. 113: 1323-1328.
6. Chang CW , Laird DA , Mausbach MJ and Hurburgh Jr. CR. 2001. Near-infrared reflectance spectroscopyprincipal components regression analysis of soil properties. Soil Sci. Soc. of Am. J. 65: 480-490.
7. Gautier H , Diakou-Verdin V , Bénard C , Reich M , Buret M , Bourgaud F and Génard M. 2008. How does tomato quality (sugar, acid, and nutritional quality) vary with ripening stage, temperature, and irradiance. J. of Agric. and Food Chem. 56: 1241-1250.
8. Golic M and Walsh KB. 2006. Robustness of calibration models based on near infrared spectroscopy for the in-line grading of stone fruit for total soluble solids content. Analytica Chimica Acta. 555: 286-291.
9. Ha JS , Jin DE , Park SK , Park CH , Seung TW , Bae DW and Heo HJ. 2015. Antiamnesic effect of Actinidia arguta extract intake in a mouse model of TMT-induced learning and memory dysfunction. Evidence-Based Complementary and Alternative Medicine: v2015, Article ID 876484. pp.13.
10. Harel-Beja R , Tzuri G , Portnoy V , Lotan-Pompan M , Lev S , Cohen S and Avisar E. 2010. A genetic map of melon highly enriched with fruit quality QTLs and EST markers, including sugar and carotenoid metabolism genes. Theoretical and Appl. Genetics. 121: 511-533.
11. Hu W , Sun DW and Blasco J. 2017. Rapid monitoring 1-MCP-induced modulation of sugars accumulation in ripening ‘Hayward’ kiwifruit by Vis/NIR hyperspectral imaging. Postharvest Biol. and Technol. 125: 168-180.
12. Huh KO and Park SY. 2007. Consumers’ behavior in the purchase of well-being products by lifestyle. Korean Academic Soc. of Business Administration. 1-25.
13. Jenny G , Mark H , Anna J , Inger K , Douglas M and Roland M.2002. Evaluation of redox indicators and the use of digital scanners and spectrophotometer for quantification of microbial growth in microplates. J. of Microbiol. methods. 50: 63-73.
14. Kalaj YR , Geyer M and Herppich WB. 2018. Non- Destructive evaluation of edible coatings effects on keeping quality of european plums (Prunus domestica L.) by laser light backscattering imaging. Erwerbs-Obstbau. 1-10.
15. Kawachi N , Kikuchi K , Suzui N , Ishii S , Fujimaki S , Ishioka NS and Watabe H. 2011. Imaging of carbon translocation to fruit using carbon-11-labeled carbon dioxide and positron emission tomography. IEEE Trans. on Nuclear Sci. 58: 395-399.
16. Kim AN , Kim HJ , Chun J , Heo HJ , Kerr WL and Choi SG. 2018. Degradation kinetics of phenolic content and antioxidant activity of hardy kiwifruit (Actinidia arguta) puree at different storage temperatures. Food Sci. and Technol. 89: 535-541.
17. Kuznets S. 1955. Economic growth and income inequality. The Am. Economic Rev. 1-28.
18. Kweon G. 2012. Toward the ultimate soil survey: sensing multiple soil and landscape properties in one pass. Agron. J. 104: 1547-1557.
19. Michalik JK and Sloan NY. 1987. Automatic refractometer. US patent 4,640,616.
20. Noh SH. 2000. Assessment of internal quality in fruits using Near-Infrared Reflectance spectroscopy. The Korean Soc. of Food Sci. and Nutrition. 124-137.
21. Ohno N , Suzuki I , Oikawa S , Sato K , Miyazaki T and Yadomae T. 1984. Antitumor activity and structural characterization of glucans extracted from cultured fruit bodies of Grifola frondosa. Chem. and pharmaceutical bulletin. 32: 1142-1151.
22. Pissard A , Baeten V , Dardenne P , Dupont P and Lateur M. 2018. Use of NIR spectroscopy on fresh apples to determine the phenolic compounds and dry matter content in peel and flesh. Biotech., Agron., Soc. and Environ. 22: 3-12.
23. Rambla FJ , Garrigues S and de la Guardia M. 1997. PLS-NIR determination of total sugar, glucose, fructose and sucrose in aqueous solutions of fruit juices. Analytica Chimica Acta. 344: 41-53.
24. Rodriguez-Saona LE , Fry FS , McLaughlin MA and Calvey EM. 2001. Rapid analysis of sugars in fruit juices by FT-NIR spectroscopy. Carbohydrate Res. 336: 63-74.
25. Sohn KH , Kim BK , Kim JY , Song WJ , Kang HR , Park HW and Min KU. 2017. Fixed food eruption caused by Actinidia arguta (hardy kiwi): a case report and literature review. Allergy, asthma & immunology res. 9: 182-184.
26. Stevens JS and Schroeder SL. 2009. Quantitative analysis of saccharides by X-ray photoelectron spectroscopy. Surface and Interface Analysis. 41: 453-462.
27. Wolpert JA , Howell GS and Mansfield TK. 1983. Sampling Vidal blanc grapes. I. Effect of training system, pruning severity, shoot exposure, shoot origin, and cluster thinning on cluster weight and fruit quality. Am. J. of Enology and Viticulture. 34: 72-76.
 오늘하루 팝업창 안보기 닫기