- Email: [email protected]
near infrared spectrometric technique for nondestructive assessment of tomato ‘Heatwave’ (Lycopersicum esculentum) quality characteristics
Journal of Food Engineering 81 (2007) 672–678 www.elsevier.com/locate/jfoodeng
Visible/near infrared spectrometric technique for nondestructive assessment of tomato ‘Heatwave’ (Lycopersicum esculentum) quality characteristics Yongni Shao a, Yong He a, Antihus H. Go´mez b, Annia G. Pereir b, Zhengjun Qiu a,*, Yun Zhang a a
College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310029, China b Agricultural Mechanization Department, Agricultural University of Havana, Cuba
Received 29 September 2006; received in revised form 24 December 2006; accepted 27 December 2006 Available online 21 January 2007
Abstract Nondestructive optical methods based on visible and near infrared reflectance spectroscopy (Vis/NIRS) have been used for estimation of physiological properties of batches of fruit and vegetable products. The objectives of this study were to evaluate the application of Vis/ NIRS in measuring the quality characteristics of tomato ‘Heatwave’ (Lycopersicum esculentum), including fruit firmness (indicated by compression force (Fc) and puncture force (Fp)), soluble solids content (SSC) and acidity (pH). Reflectance (R) determinations in the 350–2500 nm range were done on 200 tomato samples separated randomly into two groups: 170 fruit for method calibration and the remaining 30 for predictions of quality. The best calibration model for each characteristic was obtained by principal component regression (PCR) and partial least square regression (PLS) analysis. Excellent prediction performance was achieved for each tomato quality characteristic. The correlation coefficient and standard error of prediction to soluble solids content were 0.90 and 0.19°Brix, respectively. The corresponding values for pH, Fc and Fp were 0.83 and 0.09, 0.81 and 16.017 N, and 0.83 and 1.18 N, respectively. Comparatively, the model had significantly greater accuracy in determining SSC. These results suggest that Vis/NIRS measurements in the full spectral range (400–2350 nm) could be used to assess certain tomato quality, which can support further investigation into developing wider calibration from more varied growing condition or wider range of varieties. Ó 2007 Elsevier Ltd. All rights reserved. Keywords: Visible/NIR spectroscopy; Nondestructive technique; Firmness; Sugar content; Acidity; Tomato
1. Introduction Fruits are sorted manually or automatically on the basis of size, color, and surface defects such as bruises, chilling injuries, and scalds. However, total soluble solids content, sugar content, juice acidity and firmness are more important quality attributes of fruit quality. Most instrumental techniques measuring these properties are destructive and involve a considerable amount of manual work. Nondestructive techniques measuring fruit quality such as meth-
*
Corresponding author. Tel./fax: +86 571 86971143. E-mail addresses: [email protected], [email protected] (Z. Qiu).
0260-8774/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.jfoodeng.2006.12.026
ods based on visible and near infrared reflectance spectroscopy (Vis/NIRS, 350–2500 nm wavelength) are needed. The Vis/NIRS technique requires limited sample preparation and achieves high analysis speed. Properly calibrated instruments can be used for days or months and have been utilized for analyzing large batches of fruit products (Batten, 1998). Tomato ‘Heatwave’ (L. esculentum) is one of the most important fruit in the agriculture markets of China and favored by many people. However, the different varieties of tomato are of different taste and quality. Both the appearances (shape, color, size, firmness, etc) and the interior qualities (soluble solids content, acidity and the vitamin content, etc) are the aspects which can be used as the
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
quality criterion of tomato (Artes, Conesa, Hernandez, & Gil, 1999; Arazuri, Jaren, Arana, & Perez de Ciriza, 2007), thereinto firmness (compression force (Fc) and puncture force (Fp)) (Diezma-lglesias, Valero, Garcia-Ramos, & Ruiz-Altisent, 2006; Arazuri et al., 2007; Mwithiga, Mukolwe, Shitanda, & Karanja, 2007), soluble solids content (SSC) and acidity (pH) are the most important evaluation criterion which affects the consumers’ appreciation for selection. Most of the methods to measure these qualities are based on complex processing of samples, the most expensive chemical reagents and so on. Nondestructive optical methods based on visible/NIR spectrometry have been evaluated for assessing the above quality of batches of fruits and vegetable such as mandarin (Gomez, He, & Pereira, 2006; McGlone, Fraser, Jordan, & Ku¨nnemeyer, 2003), mango (Saranwong, Sornsrivichai, & Kawano, 2004; Schmilovitch, Mizrach, Hoffman, Egozi, & Fuchs, 2000), kiwifruit (McGlone, Jordan, Seelye, & Martinsen, 2002; Osborne & Ku¨nnemeyer, 1999), peach (Slaughter, 1995; Peiris, Dull, Leffler, & Kays, 1997; Peiris, Dull, Leffler, & Kays, 1998, 1998b; Golic & Walsh, 2006), and apple (Lammertyn, Nicolai, Ooms, Smedt, & Baerdemaeker, 1998; Lu, Guyer, & Beaudry, 2000; McGlone, Jordan, & Martinsen, 2002; He, Li, & Shao, 2006). Also, on tomato (Slaughter, Barrett, & Boersig, 1996; Peiris et al., 1998; Jha & Matsuoka, 2004; Pedro & Ferreira, 2005), there have some publications, but all of them were focused on SSC or acidity, and no research on firmness analysis, which was also important to tomato chosen. The objectives of this study were to evaluate the use of Vis/NIR spectroscopy in measuring the quality characteristics of tomato ‘Heatwave’ (L. esculentum), and to establish the relationship between the Vis/NIR spectral measurements and the major physiological properties of tomatoes, including fruit firmness, soluble solids content (SSC) and acidity (pH). Multivariate calibration techniques, such as principal component analysis (PCA) (He, Feng, Deng, & Li, 2006), principal component regression (PCR) (Wold, Esbensen, & Geladi, 1987) and partial least squares (PLS) (Haalan & Thomas, 1988; Gao & Ren, 2005) were used for data statistical analyses and the construction of prediction models for each quality characteristic.
673
study and individually numbered. As the fruits were harvested from different plants, the experiment design was completely randomized with each tomato as an experimental unit. 2.2. Spectra acquisition A Lowell pro-lamp interior light source (Assembly/ 128930) with the Lowell pro-lamp 14.5 V Bulb/128690 tungsten halogen made in China, which could be used both in the visible and near infrared regions, was placed at a distance of 300 mm from the fruit surface (Fig. 1). With the bifurcated optical configuration, the light was guided to the sample by the source fibers and received from the sample by the detector fibers. The angle between the incident light (light source) and the detector fiber (pistol grip unit 6307) was 45° (Fig. 1). The fibers had a 4 mm2 active surface with an 8° angle, and were placed at a height of 100 mm from the fruit surface (Fig. 1). A 100 mm2 TeflonÒ disk was used as the optical reference standard for the system, since Teflon had a very high reflectance and its lightscattering characteristics were similar to those of the samples. The reflectance (R) was calculated by comparing the near infrared energy reflected from the sample with the standard reference. Three reflection measurements (350–2500 nm) were taken at three equidistant positions around the equator (approximately 120°) of each tomato, using a spectrophotometer (FieldSpec Pro FR (350–2500 nm)/A110070, Analytical Spectral Devices, Inc. (ASD)) and the RS2 software for WindowsÒ (Fig. 1). To each reflectance spectrum, the scan number was set 10 at exactly the same position and thus the total scan number for each sample was 30. All spectral measurements, including that of the Teflon standard measured prior to obtaining each fruit reflection spectrum, were recorded as an average of 30 scans. The signals were preprocessed using ViewSpec Pro V2.14 (Analytical
2. Materials and methods 2.1. Sample preparation Tomato ‘Heatwave’ (L. esculentum) were hand harvested from the same greenhouse (November 29, 2004, Hangzhou, China). Size measurements at right angles to obtain two equatorial diameters at 90° and of the line reaching from the stem to the fruit blossom were made on all harvested tomatoes using a digital caliper (0– 150 ± 0.01 mm, Mitutoyo, UK). These values were averaged to obtain a tomato size value. Two hundred samples selected on the basis of color uniformity (between pink and light red) and size (78 ± 2.5 mm) were selected for this
Fig. 1. Vis/NIRS experimental instrument setup for each tomato: the positions of the light source and the detector fiber from the sample fruit and the three equidistant angles around the equator of each tomato.
674
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
Spectral Device, Inc., Boulder, CO 80301). Average spectra reflectance (R) for each tomato was obtained as the average of spectra collected for three positions around the fruit equator (Fig. 2). Optical system limitations and the negative influences from environmental light, significant scattering was observed at the beginning and end of the spectral data. To improve optical measurement accuracy, the region corresponding to the first 50 and the last 150 nm were not used in the analysis which was based on the 400–2350 nm region. 2.3. Tomato quality parameters measurement In order to assess the real quality characteristics of fruit at harvest time. Fruit firmness by compression and Magness–Taylor puncture tests, soluble solids content (SSC) and acidity (pH) were determined as follows. Compression tests were conducted on a Universal Testing Machine (Model 5543 Single Column, Instron Corp., Canton MA. USA) using parallel plates. Fruit firmness was quantified as the maximum compression force (Fc) required compressing the fruit diameter by 3% and recorded at a strain rate of 0.00016 m/s (10 mm/min). Fc values were determined on three positions along the equator approximately 120° apart and perpendicular to the stem-bottom axis. In the Magness–Taylor puncture test, the fruit was supported by a 75 mm diameter cradle (about the same as the fruit) with a hemispherical depression in the center of an aluminum plate. A 500 N load cell and a cylindrical plunger (6 mm diameter) at a 0.00016 m/s (10 mm/min) crosshead speed was used for tests on the same three fruit locations previously described and recorded using a strip chart at 5 kgf full scale. The maximum puncture force (Fp) defined as the peak force required to cause fruit rupture was measured. The reference measurement in all cases was 0.05 N. The pH and soluble solids content (SSC) of the juice samples were determined after compression and puncture tests, using the pH meter (model 320-s) manufactured by Mettler Toledo Scale Company Shanghai corp., Switzerland and the digital refractometer WYT-J 0-32°Brix, Beijing, China, respectively. All of these measurements were performed immediately after Vis/NIRS measurements.
Fig. 2. Original reflectance spectra for one tomato (350–2500 nm) (mean value of 30 scans).
2.4. Spectra processing The effects on the prediction of the calibration models for two data preprocessing alternatives were evaluated as follows. A moving average was used to decrease the wavelength resolution of the spectra (1.4 and 2 nm in the 350– 1000 nm and 1000–2500 nm range, respectively) n times for n subsequent wavelengths by averaging reflectance values, resulting in a decrease of resolution of n times. As a result, the number of measurement points of each spectrum was reduced by 5, 9, 13, 17, 21 and 27 segment sizes, respectively. The prediction results of the calibration models based on the different smoothing factor (segment size) were compared, and the optimal segment size was selected to establish the optimal models for tomato quality evaluation. A second data preprocessing alternative was multiplicative scatter correction (MSC) to correct additive and multiplicative effects in the spectra. Due to fruit flesh light scattering, the light did not always travel the same distance from the sample under analysis. A longer light traveling path resulted in lower relative reflectance values since less light were detected causing a parallel translation of the spectra (Martens & Naes, 1987). However, this kind of variation was of no use for the calibration models and can be eliminated by MSC. All fruit reflectance measurements were transformed to absorbance (log (1/R)) values to obtain linear correlations of the NIR values with the concentration of estimated constituents. The transformed values and four quality parameters previously defined were used for statistical multivariate calibration analysis (Unscrambler V8.0.5, CAMO Inc., Oslo, Norway). Before calibration, the spectra variation was analyzed by principal component analysis (PCA) to eliminate defective spectra outliers. Principal component regression (PCR) and partial least squares (PLS) (Haalan & Thomas, 1988) were combined to build the prediction models. Multilinear regression (MLR) was not recommended due to the collinearity between adjacent wavelengths (Lammertyn et al., 1998). Also, PLS and PCR described the original data more efficiently as they were transformed into a set of orthogonal variables. For each fruit quality parameter, different calibration models were calculated depending on the results of PCR or PLS taken into consideration spectra preprocessing and the number of factors (latent variables). The following two criteria were considered when selecting the most desirable model for each tomato quality parameter (Cen, He, & Huang, 2006; Gomez et al., 2006). First, the quality of the calibration model was quantified by the standard error of calibration (SEC), the standard error of prediction (SEP) and the correlation coefficient (r) between the predicted and the measured parameters. Acceptable models should have lower SEC and SEP, high correlation coefficients and small differences between SEC and SEP. Large differences indicate that too many latent variables are introduced in the model and that data noises is also being modeled. SEC and SEP are, respectively, defined as
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678 Ic X
SEC ¼
!1=2 2
ð^y i y i Þ =ðI c 1Þ
ð1Þ
i¼1 Ip X 2 ð^y i y i biasÞ =ðI p 1Þ
SEP ¼
!1=2 ð2Þ
i¼1
where ^y i is the predicted value of each observation and yi is the measured value; Ic is the observation number in the calibration set and Ip is the observation number in the validation set; and bias is the systematic differences between predictions and measurements calculated as bias ¼
Ip X
ð^y i y i Þ=I p
ð3Þ
i¼1
Second, a relative low number of latent variables (LV) are generally desirable to avoid modeling noise signals. The minimum plot of the root mean squared error of prediction (RMSEP) of the parameters in question against the number of latent variables was used to determine the optimal number of latent variables. The correct number of regression factors for the PLS and PCR models was determined according to the minimum root mean square error of cross validation. More variables would result in ‘‘overfitted” models, while fewer would produce ‘‘underfitted” ones. The RMSEP is defined as RMSEP ¼
Ip 1 X 2 ð^y i y i Þ I p i¼1
ð4Þ
3. Results and discussion
tial to estimate not only the component concentrations but chemical and physical properties from the infrared spectra. Results obtained agreed with conclusions reported by Haalan and Thomas (1988) and Lammertyn et al. (1998) who adopted similar approaches to choose the optimal model for prediction of apple quality characteristics. Table 2 showed the summary statistics for all samples selected in each data set. When building the calibration model for each parameter, some samples were taken out, which may cause by the instrument once in a while or may by the man-made error. Two samples for SSC, one sample for compression and one sample for puncture forces, respectively, were left out, due to their potential bad influences over the models, which were indicated during PCA pre-analysis. 3.2. Prediction of individual quality factors PLS predictions for soluble solids content, acidity, compression and puncture forces can be summarized as scatter plots (Fig. 3) with the ordinate and abscissa axes represent-
Table 2 Statistical analysis of the calibration and prediction sample sets, i.e., the data ranges, means and standard deviation (S.D.) Characteristic
Item
Calibration 98
Prediction 30
SSC (°Brix)
No. Range Mean S.D
98 2.02–4.82 3.68 0.39
30 2.14–4.62 3.54 0.46
pH
No. Range Mean S.D
100 4.03–4.78 4.39 0.164
30 4.04–4.78 4.37 0.209
Fc (N)
No. Range Mean S.D
99 5.71–85.8 37.02 26.45
30 6.04–85.6 39.34 24.36
Fp (N)
No. Range Mean S.D
99 2.58–30.2 14.69 7.65
30 5.64–29.54 15.71 7.82
3.1. Selection of the optimal models Several pretreatments tested to find an acceptable model for each quality factor using the PLS and PCR approaches (Table 1), showed that the calibration methods influenced the results as well. The PLS models were slightly better than the PCR ones because PLS models did not include latent variables that were less important to describe the variance of the quality parameter (Jong, 1993). Moreover, it should be noted that PCR and PLS both had the poten-
675
Table 1 Results of calibration and full-cross validation of the optimal models by PLS and PCR (spectrum range 400–2350 nm) Parameter
Gapsize
Method
Calibration
Cross-validation
LV
r
SEC
RMSEC
r
SEP
RMSEP
SSC
13 13
PLS PCR
3 5
0.95 0.92
0.11 0.14
0.11 0.14
0.91 0.86
0.01 0.19
0.16 0.19
pH
9 9
PLS PCR
4 7
0.92 0.90
0.07 0.07
0.07 0.07
0.85 0.83
0.08 0.09
0.09 0.09
Fc
13 17
PLS PCR
6 7
0.88 0.87
1.18 1.21
1.17 1.2
0.81 0.82
1.50 1.45
1.48 1.43
Fp
21 17
PLS PCR
3 5
0.92 0.91
0.79 0.83
0.78 0.82
0.87 0.87
0.98 1.00
0.97 0.99
676
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
Fig. 3. Vis/NIRS prediction results of the established PLS models for each tomato physiological parameter, including soluble solids content (SSC), acidity (pH), compression force (Fc) and puncture forces (Fp).
ing predicted and measured fitted values, respectively. The prediction parameters, including correlation coefficient, SEP, and RMSEP for each characteristic, implied if the established prediction models were reasonable and if the prediction performances of the models were optimal. 3.3. Soluble solids content (SSC) The calibration correlation between the NIR measurements and the SSC was as high as 0.95, with the standard error of calibration (SEC) 0.11°Brix (Table 1). When the model was used to predict the other 30 samples, the prediction results were also desirable, with the correlation coefficient between the measured and the predicted (r) 0.90, the standard error of prediction (SEP) 0.19°Brix with a bias 0.33°Brix (Fig. 3). The PLS model appeared to be robust since only three factors (LVs) were used in the calibration model (Table 1). The regression coefficients of SSC obtained in this research were slightly superior to those obtained by Lammertyn et al. (1998) in apples with r = 0.82 and SEP = 0.6; by Peirs, Lammertyn, Ooms, and Nicolaı¨ (2000) in different apples varieties with r = 0.730.89; by Slaughter et al. (1996) in tomatoes with r = 0.89 and SEP = 0.33; and by Lu (2001) in the cherries Sam variety with r = 0.89 and SEP = 0.65. However, better results had been
achieved by Lu (2001) in cherries Hedelfinger variety with r = 0.97 and SEP = 0.71; by Kawano, Watanabe, and Iwamoto (1992) in peaches with r = 0.989 and SEP = 0.32. Osborne and Ku¨nnemeyer (1999), McGlone et al. (2003) had studied the kiwifruits and mandarins, respectively, and obtained the results with RMSEP = 0.27°Brix and RMSEP = 0.32°Brix, which were both minor than 0.38°Brix in this paper (Table 1) and 0.5°Brix in apples by McGlone and Kawano (1998). Peiris et al. (1997, 1998) obtained in peaches r = 0.57 and SEP = 0.74% using PLS validation. And for comparison with PLS, they also adopted neural networks and the results turned out to be better with bias = 0.03%, SEP = 0.52% and r = 0.69. However, both the above prediction results could be considered poor. 3.4. Acidity (pH) The calibration correlation between the NIR measurements and the pH of tomatoes was with r = 0.92 and SEC = 0.07 (Table 1). When the calibrated model was applied to the prediction set (30 samples), the results were applicable with r = 0.83, SEP = 0.096 and bias = 0.234 (Fig. 3). The PLS model appeared to be acceptable since four factors (LVs) were used in the calibration model (Table 1). Lammertyn et al. (1998) obtained in apples a
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
regression coefficient (r) 0.93 and SEP = 0.068, both of which were superior to the results in this research. 3.5. Compression force (Fc) In the case of compression force, a good regression coefficient was obtained in the calibration set, with r = 0.88 and SEC = 1.18 N (Table 1). When the model was used to predict the remaining 30 samples, the results were r = 0.81, SEP = 16.017 N and bias = 1.284 N (Fig. 3). The PLS model appeared to be acceptable due to the six factors (LVs) used in the calibration model (Table 1). In spite of the good correlation between the predicted and the measured compression forces, the relatively high SEP = 16.017 N might be the result of the larger standard deviation (S.D) of the prediction data set 24.36 N (Table 2). In literature, little information was found concerning NIR spectroscopy prediction models for compression force of 3% fruit high equator. 3.6. Puncture force (Fp) The calibration correlation between the NIR measurements and the puncture forces was with r = 0.92 and SEC = 0.79 N (Table 1). When the model was used to predict the remaining 30 samples, reasonable results were obtained, with r = 0.84, SEP = 1.18 N, bias = 1.938 N (Fig. 3). And three factors (LVs) were used in the calibration model (Table 1). The regression coefficients obtained in the puncture force were superior to those obtained by Lammertyn et al. (1998) in apples with a poor correlation coefficient (r) from 0.73 to 0.75, and by Lu (2001) in cherries Sam variety with r = 0.65 and SEP = 0.44 N and similarly in cherries Hedelfinger variety with r = 0.80 and SEP = 0.55 N. The RMSEP = 2.24 N was also superior to those reported in the Vis/NIRS literature on apples: RMSEP = 7.5 N in McGlone et al. (2002). In the above discussion of the PLS prediction results, no consideration was given to the contributions of the individual wavelengths to the prediction results. This was because the PLS method first applied linear transform to the entire individual wavelength data. As a result, it was often difficult to ascertain how individual wavelengths were directly related to the quantities to be predicted. However, it would be helpful to examine how sugar content, acidity or firmness was simply related to individual wavelengths so that a better understanding of their correlated NIRS might be achieved. The results suggest that Vis/NIRS measurements in the full spectral range (400–2350 nm) could be used to assess certain kind of tomato quality. In the paper, only one variety of the tomato was investigated to establish the relationship between the Vis/NIR spectral measurements and the major physiological properties, though it combined the exterior quality (mechanical characteristic) and interior quality (soluble solids content and acidity) of tomato,
677
and the results were satisfying, the stability and practicability of the model need improvement. In the further study, we should investigate into developing wider calibration from more varied growing condition or wider range of varieties of tomatoes. 4. Conclusions The advantages of NIR spectroscopy were obvious: it was a feasibly quick and nondestructive technique for measurement and the internal quality characteristics could be measured repeatedly for the same sample. The results in this paper indicated that it was possible to utilize this nondestructive technique to measure the physiological quality characteristics of certain tomatoes, with both exterior quality (mechanical characteristic) and interior quality (soluble solids content and acidity). The multivariate calibration methods PCR and PLS both had the potential to estimate the component concentrations, the chemical and physical properties of the tomatoes from their infrared spectra. By means of PLS, correlations were established between the absorbance spectra and the quality parameters of tomatoes. In the SSC model, r = 0.90 and SEP = 0.19°Brix with three factors; in the pH model, r = 0.83 and SEP = 0.09 with four factors; in the compression force model, r = 0.81 and SEP = 16.017 N with six factors, and in the puncture force model r = 0.83 and SEP = 1.18 N with three factors. All results showed better prediction performances of the established PLS models for each tomato quality characteristics. Comparatively, the Vis/NIR spectroscopy technique had significantly greater accuracy during the determination of SSC. As to the data preprocessing methods, they influenced the prediction performances of the models. In general, the models based on spectra preprocessed by MSC performed slightly better than those without MSC. For the other spectral data pretreatment, i.e., smoothing by moving average, the different segment size (5, 9, 13, 17, 21 and 27 times) applied during the initial spectral pretreatment in this paper had slight influences on the established prediction models. In the paper, it combined the exterior quality (mechanical characteristic) and interior quality (soluble solids content and acidity) of tomatoes, and the results were satisfying, the stability and practicability of the model need improvement next. In the further study, we should investigate into developing wider calibration from more varied growing condition or wider range of varieties of tomatoes. Acknowledgements This study was supported by National Science and Technology support program (2006BAD10A04) the Teaching and Research Award Program for Outstanding Young Teachers in Higher Education Institutions of MOE, PRC, Natural Science Foundation of China (Project No: 30671213), Specialized Research Fund for the Doctoral
678
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
Program of Higher Education (Project No: 20040335034) and Natural Science Foundation of Zhejiang Province, China (Project No: RC02067). References Arazuri, S., Jaren, C., Arana, J. I., & Perez de Ciriza, J. J. (2007). Influence of mechanical harvest on the physical properties of processing tomato (Lycopersicon esculentum Mill.). Journal of Food Engineering, 80, 190–198. Artes, F., Conesa, M. A., Hernandez, S., & Gil, M. I. (1999). Keeping quality of fresh-cut tomato. Postharvest Biology and Technology, 17, 153–162. Batten, G. D. (1998). Plant analysis using near infrared reflectance spectroscopy: the potential and the limitations. Australian Journal of Experimental Agriculture, 38, 697–706. Cen, H. Y., He, Y., & Huang, M. (2006). Fast measurement of soluble solids contents and acidity in orange juice using chemometrics and Vis/ NIRS. Journal of Agricultural and Food Chemistry, 54, 7437–7443. Diezma-lglesias, B., Valero, C., Garcia-Ramos, F. J., & Ruiz-Altisent, M. (2006). Monitoring of firmness evolution of peaches during storage by combing acoustic and impact methods. Journal of Food Engineering, 77, 926–935. Gao, L., & Ren, S. X. (2005). Simultaneous spectrophotometric determination of four metals by two kinds of partial least squares methods. Spectrochimica Acta, Part A, 61, 3013–3019. Golic, M., & Walsh, K. B. (2006). Robustness of calibration models based on near infrared spectroscopy for the in-line grading of stonefruit for total soluble solids content. Analytica Chimica Acta, 555, 286–291. Gomez, A. H., He, Y., & Pereira, A. G. (2006). Non-destructive measurement of acidity, soluble solids and firmness of Satsuma mandarin using Vis/NIR-spectroscopy techniques. Journal of Food Engineering, 77, 313–319. Haalan, D., & Thomas, E. (1988). Partial Least Squares method for spectral analysis. 1. Relation to other quantitative calibration methods and the extraction of qualitative information. Analytical Chemistry, 60, 1193–1202. He, Y., Feng, S. J., Deng, X. F., & Li, X. L. (2006). Study on lossless discrimination of varieties of yogurt using the visible/NIR-spectroscopy. Food Research International, 39, 645–650. He, Y., Li, X. L., & Shao, Y. N. (2006). Discrimination of varieties of apple using near infrared spectra based on principal component analysis and artificial neural network model. Spectroscopy and Spectral Analysis, 26, 850–853. Jha, S., & Matsuoka, T. (2004). Non-destructive determination of acid– brix ratio of tomato juice using near infrared spectroscopy. International Journal of Food Science and Technology, 39, 425. Jong, S. (1993). PLS fits closer than PCR. Journal of Chemometrics, 7, 551–557. Kawano, S., Watanabe, H., & Iwamoto, M. (1992). Determination of sugar content in intact peaches by near infrared spectroscopy with fiber optics in interactance mode. Journal of Japanese Society for Horticultural Science, 61, 445–451. Lammertyn, J., Nicolai, B., Ooms, K., Smedt, V., & Baerdemaeker, J. (1998). Non-destructive measurement of acidity, soluble solids, and firmness of Jonagold apples using NIR-Spectroscopy. Transactions of the ASAE, 41, 1086–1094. Lu, R. (2001). Predicting firmness and sugar content of sweet cherries using near-infrared diffuse reflectance spectroscopy. Transactions of the ASAE, 44, 1265–1271.
Lu, R., Guyer, D., & Beaudry, R. (2000). Determination of firmness and sugar content of apple using NIR diffuse reflectance. Journal of Texture Studies, 31, 615–630. Martens, H., & Naes, T. (1987). Multivariate calibration by data compression. In P. Williams, K. Norris, & S. Paul (Eds.), Near-infrared Technology in the Agricultural and Food Industries (pp. 123–139). St. Paul, Minnesota: American Association of Cereal Chemists. McGlone, V. A., Fraser, D. G., Jordan, R. B., & Ku¨nnemeyer, R. (2003). Internal quality assessment of mandarin fruit by vis/NIR spectroscopy. Journal of Near Infrared Spectroscopy, 11, 323–332. McGlone, V. A., Jordan, R. B., & Martinsen, P. J. (2002). Vis/NIR estimation at harvest of pre- and post-storage quality indices for ‘Royal Gala’ apple. Postharvest Biology and Technology, 25, 135–144. McGlone, V. A., Jordan, R. B., Seelye, R., & Martinsen, P. J. (2002). Comparing density and NIR methods for measurement of Kiwifruit dry matter and soluble solids content. Postharvest Biology and Technology, 26, 191–198. McGlone, V., & Kawano, S. (1998). Firmness, dry-matter and solublesolids assessment of postharvest kiwifruit by NIR spectroscopy. Postharvest Biology and Technology, 13, 131–141. Mwithiga, G., Mukolwe, M. I., Shitanda, D., & Karanja, P. N. (2007). Evaluation of the effect of ripening on the sensory quality and properties of tamarillo (Cyphomandra betaceae) fruits. Journal of Food Engineering, 79, 117–123. Osborne, S., & Ku¨nnemeyer, R. (1999). A low-cost system for the grading of kiwifruit. Journal of Near Infrared Spectroscopy, 7, 9–15. Pedro, A. M. K., & Ferreira, M. M. C. (2005). Nondestructive determination of solids and carotenoids in tomato products by nearinfrared spectroscopy and multivariate calibration. Analytical Chemistry, 77, 2505–2511. Peiris, K. H. S., Dull, G. G., Leffler, R. G., & Kays, S. J. (1998). Nearinfrared spectrometric method for nondestructive determination of soluble solids content of peaches. Journal of the American Society for Horticulture Science, 123, 898–905. Peiris, K. H. S., Dull, G. G., Leffler, R. G., & Kays, S. J. (1998b). Nearinfrared spectrometric method for nondestructive determination of soluble solids content in processing tomatoes. Journal of the American Society for Horticulture Science, 123, 1089–1093. Peiris, K., Dull, G., Leffler, R., & Kays, S. (1997). Nondestructive determination of soluble solids content of peaches by near infrared spectroscopy. International Conference of Sensors for Nondestructive Testing: measuring the quality of fresh fruit and vegetables. NRAES: Ithaca, NY, pp.77–87. Peirs, A., Lammertyn, J., Ooms, K., & Nicolaı¨, B. M. (2000). Prediction of the optimal picking date of different apple cultivars by means of VIS: NIR-spectroscopy. Postharvest Biology and Technology, 21, 189–199. Saranwong, S., Sornsrivichai, J., & Kawano, S. (2004). Prediction of ripestage eating quality of mango fruit from its harvest quality measured nondestructively by near infrared spectroscopy. Postharvest Biology and Technology, 31, 137–145. Schmilovitch, Z., Mizrach, A., Hoffman, A., Egozi, H., & Fuchs, Y. (2000). Determination of mango physiological indices by nearinfrared spectrometry. Postharvest Biology and Technology, 19, 245–252. Slaughter, D. (1995). Nondestructive determination of internal quality in peaches and nectarines. Transactions of the ASAE, 38, 617–623. Slaughter, D., Barrett, D., & Boersig, M. (1996). Nondestructive determination of soluble solids in tomatoes using near infrared spectroscopy. Journal of Food Science, 61, 695–697. Wold, S., Esbensen, K., & Geladi, P. (1987). Principal component analysis. Chemometric and intelligent Laboratory System, 2, 37–52.
Visible/near infrared spectrometric technique for nondestructive assessment of tomato ‘Heatwave’ (Lycopersicum esculentum) quality characteristics Yongni Shao a, Yong He a, Antihus H. Go´mez b, Annia G. Pereir b, Zhengjun Qiu a,*, Yun Zhang a a
College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310029, China b Agricultural Mechanization Department, Agricultural University of Havana, Cuba
Received 29 September 2006; received in revised form 24 December 2006; accepted 27 December 2006 Available online 21 January 2007
Abstract Nondestructive optical methods based on visible and near infrared reflectance spectroscopy (Vis/NIRS) have been used for estimation of physiological properties of batches of fruit and vegetable products. The objectives of this study were to evaluate the application of Vis/ NIRS in measuring the quality characteristics of tomato ‘Heatwave’ (Lycopersicum esculentum), including fruit firmness (indicated by compression force (Fc) and puncture force (Fp)), soluble solids content (SSC) and acidity (pH). Reflectance (R) determinations in the 350–2500 nm range were done on 200 tomato samples separated randomly into two groups: 170 fruit for method calibration and the remaining 30 for predictions of quality. The best calibration model for each characteristic was obtained by principal component regression (PCR) and partial least square regression (PLS) analysis. Excellent prediction performance was achieved for each tomato quality characteristic. The correlation coefficient and standard error of prediction to soluble solids content were 0.90 and 0.19°Brix, respectively. The corresponding values for pH, Fc and Fp were 0.83 and 0.09, 0.81 and 16.017 N, and 0.83 and 1.18 N, respectively. Comparatively, the model had significantly greater accuracy in determining SSC. These results suggest that Vis/NIRS measurements in the full spectral range (400–2350 nm) could be used to assess certain tomato quality, which can support further investigation into developing wider calibration from more varied growing condition or wider range of varieties. Ó 2007 Elsevier Ltd. All rights reserved. Keywords: Visible/NIR spectroscopy; Nondestructive technique; Firmness; Sugar content; Acidity; Tomato
1. Introduction Fruits are sorted manually or automatically on the basis of size, color, and surface defects such as bruises, chilling injuries, and scalds. However, total soluble solids content, sugar content, juice acidity and firmness are more important quality attributes of fruit quality. Most instrumental techniques measuring these properties are destructive and involve a considerable amount of manual work. Nondestructive techniques measuring fruit quality such as meth-
*
Corresponding author. Tel./fax: +86 571 86971143. E-mail addresses: [email protected], [email protected] (Z. Qiu).
0260-8774/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.jfoodeng.2006.12.026
ods based on visible and near infrared reflectance spectroscopy (Vis/NIRS, 350–2500 nm wavelength) are needed. The Vis/NIRS technique requires limited sample preparation and achieves high analysis speed. Properly calibrated instruments can be used for days or months and have been utilized for analyzing large batches of fruit products (Batten, 1998). Tomato ‘Heatwave’ (L. esculentum) is one of the most important fruit in the agriculture markets of China and favored by many people. However, the different varieties of tomato are of different taste and quality. Both the appearances (shape, color, size, firmness, etc) and the interior qualities (soluble solids content, acidity and the vitamin content, etc) are the aspects which can be used as the
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
quality criterion of tomato (Artes, Conesa, Hernandez, & Gil, 1999; Arazuri, Jaren, Arana, & Perez de Ciriza, 2007), thereinto firmness (compression force (Fc) and puncture force (Fp)) (Diezma-lglesias, Valero, Garcia-Ramos, & Ruiz-Altisent, 2006; Arazuri et al., 2007; Mwithiga, Mukolwe, Shitanda, & Karanja, 2007), soluble solids content (SSC) and acidity (pH) are the most important evaluation criterion which affects the consumers’ appreciation for selection. Most of the methods to measure these qualities are based on complex processing of samples, the most expensive chemical reagents and so on. Nondestructive optical methods based on visible/NIR spectrometry have been evaluated for assessing the above quality of batches of fruits and vegetable such as mandarin (Gomez, He, & Pereira, 2006; McGlone, Fraser, Jordan, & Ku¨nnemeyer, 2003), mango (Saranwong, Sornsrivichai, & Kawano, 2004; Schmilovitch, Mizrach, Hoffman, Egozi, & Fuchs, 2000), kiwifruit (McGlone, Jordan, Seelye, & Martinsen, 2002; Osborne & Ku¨nnemeyer, 1999), peach (Slaughter, 1995; Peiris, Dull, Leffler, & Kays, 1997; Peiris, Dull, Leffler, & Kays, 1998, 1998b; Golic & Walsh, 2006), and apple (Lammertyn, Nicolai, Ooms, Smedt, & Baerdemaeker, 1998; Lu, Guyer, & Beaudry, 2000; McGlone, Jordan, & Martinsen, 2002; He, Li, & Shao, 2006). Also, on tomato (Slaughter, Barrett, & Boersig, 1996; Peiris et al., 1998; Jha & Matsuoka, 2004; Pedro & Ferreira, 2005), there have some publications, but all of them were focused on SSC or acidity, and no research on firmness analysis, which was also important to tomato chosen. The objectives of this study were to evaluate the use of Vis/NIR spectroscopy in measuring the quality characteristics of tomato ‘Heatwave’ (L. esculentum), and to establish the relationship between the Vis/NIR spectral measurements and the major physiological properties of tomatoes, including fruit firmness, soluble solids content (SSC) and acidity (pH). Multivariate calibration techniques, such as principal component analysis (PCA) (He, Feng, Deng, & Li, 2006), principal component regression (PCR) (Wold, Esbensen, & Geladi, 1987) and partial least squares (PLS) (Haalan & Thomas, 1988; Gao & Ren, 2005) were used for data statistical analyses and the construction of prediction models for each quality characteristic.
673
study and individually numbered. As the fruits were harvested from different plants, the experiment design was completely randomized with each tomato as an experimental unit. 2.2. Spectra acquisition A Lowell pro-lamp interior light source (Assembly/ 128930) with the Lowell pro-lamp 14.5 V Bulb/128690 tungsten halogen made in China, which could be used both in the visible and near infrared regions, was placed at a distance of 300 mm from the fruit surface (Fig. 1). With the bifurcated optical configuration, the light was guided to the sample by the source fibers and received from the sample by the detector fibers. The angle between the incident light (light source) and the detector fiber (pistol grip unit 6307) was 45° (Fig. 1). The fibers had a 4 mm2 active surface with an 8° angle, and were placed at a height of 100 mm from the fruit surface (Fig. 1). A 100 mm2 TeflonÒ disk was used as the optical reference standard for the system, since Teflon had a very high reflectance and its lightscattering characteristics were similar to those of the samples. The reflectance (R) was calculated by comparing the near infrared energy reflected from the sample with the standard reference. Three reflection measurements (350–2500 nm) were taken at three equidistant positions around the equator (approximately 120°) of each tomato, using a spectrophotometer (FieldSpec Pro FR (350–2500 nm)/A110070, Analytical Spectral Devices, Inc. (ASD)) and the RS2 software for WindowsÒ (Fig. 1). To each reflectance spectrum, the scan number was set 10 at exactly the same position and thus the total scan number for each sample was 30. All spectral measurements, including that of the Teflon standard measured prior to obtaining each fruit reflection spectrum, were recorded as an average of 30 scans. The signals were preprocessed using ViewSpec Pro V2.14 (Analytical
2. Materials and methods 2.1. Sample preparation Tomato ‘Heatwave’ (L. esculentum) were hand harvested from the same greenhouse (November 29, 2004, Hangzhou, China). Size measurements at right angles to obtain two equatorial diameters at 90° and of the line reaching from the stem to the fruit blossom were made on all harvested tomatoes using a digital caliper (0– 150 ± 0.01 mm, Mitutoyo, UK). These values were averaged to obtain a tomato size value. Two hundred samples selected on the basis of color uniformity (between pink and light red) and size (78 ± 2.5 mm) were selected for this
Fig. 1. Vis/NIRS experimental instrument setup for each tomato: the positions of the light source and the detector fiber from the sample fruit and the three equidistant angles around the equator of each tomato.
674
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
Spectral Device, Inc., Boulder, CO 80301). Average spectra reflectance (R) for each tomato was obtained as the average of spectra collected for three positions around the fruit equator (Fig. 2). Optical system limitations and the negative influences from environmental light, significant scattering was observed at the beginning and end of the spectral data. To improve optical measurement accuracy, the region corresponding to the first 50 and the last 150 nm were not used in the analysis which was based on the 400–2350 nm region. 2.3. Tomato quality parameters measurement In order to assess the real quality characteristics of fruit at harvest time. Fruit firmness by compression and Magness–Taylor puncture tests, soluble solids content (SSC) and acidity (pH) were determined as follows. Compression tests were conducted on a Universal Testing Machine (Model 5543 Single Column, Instron Corp., Canton MA. USA) using parallel plates. Fruit firmness was quantified as the maximum compression force (Fc) required compressing the fruit diameter by 3% and recorded at a strain rate of 0.00016 m/s (10 mm/min). Fc values were determined on three positions along the equator approximately 120° apart and perpendicular to the stem-bottom axis. In the Magness–Taylor puncture test, the fruit was supported by a 75 mm diameter cradle (about the same as the fruit) with a hemispherical depression in the center of an aluminum plate. A 500 N load cell and a cylindrical plunger (6 mm diameter) at a 0.00016 m/s (10 mm/min) crosshead speed was used for tests on the same three fruit locations previously described and recorded using a strip chart at 5 kgf full scale. The maximum puncture force (Fp) defined as the peak force required to cause fruit rupture was measured. The reference measurement in all cases was 0.05 N. The pH and soluble solids content (SSC) of the juice samples were determined after compression and puncture tests, using the pH meter (model 320-s) manufactured by Mettler Toledo Scale Company Shanghai corp., Switzerland and the digital refractometer WYT-J 0-32°Brix, Beijing, China, respectively. All of these measurements were performed immediately after Vis/NIRS measurements.
Fig. 2. Original reflectance spectra for one tomato (350–2500 nm) (mean value of 30 scans).
2.4. Spectra processing The effects on the prediction of the calibration models for two data preprocessing alternatives were evaluated as follows. A moving average was used to decrease the wavelength resolution of the spectra (1.4 and 2 nm in the 350– 1000 nm and 1000–2500 nm range, respectively) n times for n subsequent wavelengths by averaging reflectance values, resulting in a decrease of resolution of n times. As a result, the number of measurement points of each spectrum was reduced by 5, 9, 13, 17, 21 and 27 segment sizes, respectively. The prediction results of the calibration models based on the different smoothing factor (segment size) were compared, and the optimal segment size was selected to establish the optimal models for tomato quality evaluation. A second data preprocessing alternative was multiplicative scatter correction (MSC) to correct additive and multiplicative effects in the spectra. Due to fruit flesh light scattering, the light did not always travel the same distance from the sample under analysis. A longer light traveling path resulted in lower relative reflectance values since less light were detected causing a parallel translation of the spectra (Martens & Naes, 1987). However, this kind of variation was of no use for the calibration models and can be eliminated by MSC. All fruit reflectance measurements were transformed to absorbance (log (1/R)) values to obtain linear correlations of the NIR values with the concentration of estimated constituents. The transformed values and four quality parameters previously defined were used for statistical multivariate calibration analysis (Unscrambler V8.0.5, CAMO Inc., Oslo, Norway). Before calibration, the spectra variation was analyzed by principal component analysis (PCA) to eliminate defective spectra outliers. Principal component regression (PCR) and partial least squares (PLS) (Haalan & Thomas, 1988) were combined to build the prediction models. Multilinear regression (MLR) was not recommended due to the collinearity between adjacent wavelengths (Lammertyn et al., 1998). Also, PLS and PCR described the original data more efficiently as they were transformed into a set of orthogonal variables. For each fruit quality parameter, different calibration models were calculated depending on the results of PCR or PLS taken into consideration spectra preprocessing and the number of factors (latent variables). The following two criteria were considered when selecting the most desirable model for each tomato quality parameter (Cen, He, & Huang, 2006; Gomez et al., 2006). First, the quality of the calibration model was quantified by the standard error of calibration (SEC), the standard error of prediction (SEP) and the correlation coefficient (r) between the predicted and the measured parameters. Acceptable models should have lower SEC and SEP, high correlation coefficients and small differences between SEC and SEP. Large differences indicate that too many latent variables are introduced in the model and that data noises is also being modeled. SEC and SEP are, respectively, defined as
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678 Ic X
SEC ¼
!1=2 2
ð^y i y i Þ =ðI c 1Þ
ð1Þ
i¼1 Ip X 2 ð^y i y i biasÞ =ðI p 1Þ
SEP ¼
!1=2 ð2Þ
i¼1
where ^y i is the predicted value of each observation and yi is the measured value; Ic is the observation number in the calibration set and Ip is the observation number in the validation set; and bias is the systematic differences between predictions and measurements calculated as bias ¼
Ip X
ð^y i y i Þ=I p
ð3Þ
i¼1
Second, a relative low number of latent variables (LV) are generally desirable to avoid modeling noise signals. The minimum plot of the root mean squared error of prediction (RMSEP) of the parameters in question against the number of latent variables was used to determine the optimal number of latent variables. The correct number of regression factors for the PLS and PCR models was determined according to the minimum root mean square error of cross validation. More variables would result in ‘‘overfitted” models, while fewer would produce ‘‘underfitted” ones. The RMSEP is defined as RMSEP ¼
Ip 1 X 2 ð^y i y i Þ I p i¼1
ð4Þ
3. Results and discussion
tial to estimate not only the component concentrations but chemical and physical properties from the infrared spectra. Results obtained agreed with conclusions reported by Haalan and Thomas (1988) and Lammertyn et al. (1998) who adopted similar approaches to choose the optimal model for prediction of apple quality characteristics. Table 2 showed the summary statistics for all samples selected in each data set. When building the calibration model for each parameter, some samples were taken out, which may cause by the instrument once in a while or may by the man-made error. Two samples for SSC, one sample for compression and one sample for puncture forces, respectively, were left out, due to their potential bad influences over the models, which were indicated during PCA pre-analysis. 3.2. Prediction of individual quality factors PLS predictions for soluble solids content, acidity, compression and puncture forces can be summarized as scatter plots (Fig. 3) with the ordinate and abscissa axes represent-
Table 2 Statistical analysis of the calibration and prediction sample sets, i.e., the data ranges, means and standard deviation (S.D.) Characteristic
Item
Calibration 98
Prediction 30
SSC (°Brix)
No. Range Mean S.D
98 2.02–4.82 3.68 0.39
30 2.14–4.62 3.54 0.46
pH
No. Range Mean S.D
100 4.03–4.78 4.39 0.164
30 4.04–4.78 4.37 0.209
Fc (N)
No. Range Mean S.D
99 5.71–85.8 37.02 26.45
30 6.04–85.6 39.34 24.36
Fp (N)
No. Range Mean S.D
99 2.58–30.2 14.69 7.65
30 5.64–29.54 15.71 7.82
3.1. Selection of the optimal models Several pretreatments tested to find an acceptable model for each quality factor using the PLS and PCR approaches (Table 1), showed that the calibration methods influenced the results as well. The PLS models were slightly better than the PCR ones because PLS models did not include latent variables that were less important to describe the variance of the quality parameter (Jong, 1993). Moreover, it should be noted that PCR and PLS both had the poten-
675
Table 1 Results of calibration and full-cross validation of the optimal models by PLS and PCR (spectrum range 400–2350 nm) Parameter
Gapsize
Method
Calibration
Cross-validation
LV
r
SEC
RMSEC
r
SEP
RMSEP
SSC
13 13
PLS PCR
3 5
0.95 0.92
0.11 0.14
0.11 0.14
0.91 0.86
0.01 0.19
0.16 0.19
pH
9 9
PLS PCR
4 7
0.92 0.90
0.07 0.07
0.07 0.07
0.85 0.83
0.08 0.09
0.09 0.09
Fc
13 17
PLS PCR
6 7
0.88 0.87
1.18 1.21
1.17 1.2
0.81 0.82
1.50 1.45
1.48 1.43
Fp
21 17
PLS PCR
3 5
0.92 0.91
0.79 0.83
0.78 0.82
0.87 0.87
0.98 1.00
0.97 0.99
676
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
Fig. 3. Vis/NIRS prediction results of the established PLS models for each tomato physiological parameter, including soluble solids content (SSC), acidity (pH), compression force (Fc) and puncture forces (Fp).
ing predicted and measured fitted values, respectively. The prediction parameters, including correlation coefficient, SEP, and RMSEP for each characteristic, implied if the established prediction models were reasonable and if the prediction performances of the models were optimal. 3.3. Soluble solids content (SSC) The calibration correlation between the NIR measurements and the SSC was as high as 0.95, with the standard error of calibration (SEC) 0.11°Brix (Table 1). When the model was used to predict the other 30 samples, the prediction results were also desirable, with the correlation coefficient between the measured and the predicted (r) 0.90, the standard error of prediction (SEP) 0.19°Brix with a bias 0.33°Brix (Fig. 3). The PLS model appeared to be robust since only three factors (LVs) were used in the calibration model (Table 1). The regression coefficients of SSC obtained in this research were slightly superior to those obtained by Lammertyn et al. (1998) in apples with r = 0.82 and SEP = 0.6; by Peirs, Lammertyn, Ooms, and Nicolaı¨ (2000) in different apples varieties with r = 0.730.89; by Slaughter et al. (1996) in tomatoes with r = 0.89 and SEP = 0.33; and by Lu (2001) in the cherries Sam variety with r = 0.89 and SEP = 0.65. However, better results had been
achieved by Lu (2001) in cherries Hedelfinger variety with r = 0.97 and SEP = 0.71; by Kawano, Watanabe, and Iwamoto (1992) in peaches with r = 0.989 and SEP = 0.32. Osborne and Ku¨nnemeyer (1999), McGlone et al. (2003) had studied the kiwifruits and mandarins, respectively, and obtained the results with RMSEP = 0.27°Brix and RMSEP = 0.32°Brix, which were both minor than 0.38°Brix in this paper (Table 1) and 0.5°Brix in apples by McGlone and Kawano (1998). Peiris et al. (1997, 1998) obtained in peaches r = 0.57 and SEP = 0.74% using PLS validation. And for comparison with PLS, they also adopted neural networks and the results turned out to be better with bias = 0.03%, SEP = 0.52% and r = 0.69. However, both the above prediction results could be considered poor. 3.4. Acidity (pH) The calibration correlation between the NIR measurements and the pH of tomatoes was with r = 0.92 and SEC = 0.07 (Table 1). When the calibrated model was applied to the prediction set (30 samples), the results were applicable with r = 0.83, SEP = 0.096 and bias = 0.234 (Fig. 3). The PLS model appeared to be acceptable since four factors (LVs) were used in the calibration model (Table 1). Lammertyn et al. (1998) obtained in apples a
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
regression coefficient (r) 0.93 and SEP = 0.068, both of which were superior to the results in this research. 3.5. Compression force (Fc) In the case of compression force, a good regression coefficient was obtained in the calibration set, with r = 0.88 and SEC = 1.18 N (Table 1). When the model was used to predict the remaining 30 samples, the results were r = 0.81, SEP = 16.017 N and bias = 1.284 N (Fig. 3). The PLS model appeared to be acceptable due to the six factors (LVs) used in the calibration model (Table 1). In spite of the good correlation between the predicted and the measured compression forces, the relatively high SEP = 16.017 N might be the result of the larger standard deviation (S.D) of the prediction data set 24.36 N (Table 2). In literature, little information was found concerning NIR spectroscopy prediction models for compression force of 3% fruit high equator. 3.6. Puncture force (Fp) The calibration correlation between the NIR measurements and the puncture forces was with r = 0.92 and SEC = 0.79 N (Table 1). When the model was used to predict the remaining 30 samples, reasonable results were obtained, with r = 0.84, SEP = 1.18 N, bias = 1.938 N (Fig. 3). And three factors (LVs) were used in the calibration model (Table 1). The regression coefficients obtained in the puncture force were superior to those obtained by Lammertyn et al. (1998) in apples with a poor correlation coefficient (r) from 0.73 to 0.75, and by Lu (2001) in cherries Sam variety with r = 0.65 and SEP = 0.44 N and similarly in cherries Hedelfinger variety with r = 0.80 and SEP = 0.55 N. The RMSEP = 2.24 N was also superior to those reported in the Vis/NIRS literature on apples: RMSEP = 7.5 N in McGlone et al. (2002). In the above discussion of the PLS prediction results, no consideration was given to the contributions of the individual wavelengths to the prediction results. This was because the PLS method first applied linear transform to the entire individual wavelength data. As a result, it was often difficult to ascertain how individual wavelengths were directly related to the quantities to be predicted. However, it would be helpful to examine how sugar content, acidity or firmness was simply related to individual wavelengths so that a better understanding of their correlated NIRS might be achieved. The results suggest that Vis/NIRS measurements in the full spectral range (400–2350 nm) could be used to assess certain kind of tomato quality. In the paper, only one variety of the tomato was investigated to establish the relationship between the Vis/NIR spectral measurements and the major physiological properties, though it combined the exterior quality (mechanical characteristic) and interior quality (soluble solids content and acidity) of tomato,
677
and the results were satisfying, the stability and practicability of the model need improvement. In the further study, we should investigate into developing wider calibration from more varied growing condition or wider range of varieties of tomatoes. 4. Conclusions The advantages of NIR spectroscopy were obvious: it was a feasibly quick and nondestructive technique for measurement and the internal quality characteristics could be measured repeatedly for the same sample. The results in this paper indicated that it was possible to utilize this nondestructive technique to measure the physiological quality characteristics of certain tomatoes, with both exterior quality (mechanical characteristic) and interior quality (soluble solids content and acidity). The multivariate calibration methods PCR and PLS both had the potential to estimate the component concentrations, the chemical and physical properties of the tomatoes from their infrared spectra. By means of PLS, correlations were established between the absorbance spectra and the quality parameters of tomatoes. In the SSC model, r = 0.90 and SEP = 0.19°Brix with three factors; in the pH model, r = 0.83 and SEP = 0.09 with four factors; in the compression force model, r = 0.81 and SEP = 16.017 N with six factors, and in the puncture force model r = 0.83 and SEP = 1.18 N with three factors. All results showed better prediction performances of the established PLS models for each tomato quality characteristics. Comparatively, the Vis/NIR spectroscopy technique had significantly greater accuracy during the determination of SSC. As to the data preprocessing methods, they influenced the prediction performances of the models. In general, the models based on spectra preprocessed by MSC performed slightly better than those without MSC. For the other spectral data pretreatment, i.e., smoothing by moving average, the different segment size (5, 9, 13, 17, 21 and 27 times) applied during the initial spectral pretreatment in this paper had slight influences on the established prediction models. In the paper, it combined the exterior quality (mechanical characteristic) and interior quality (soluble solids content and acidity) of tomatoes, and the results were satisfying, the stability and practicability of the model need improvement next. In the further study, we should investigate into developing wider calibration from more varied growing condition or wider range of varieties of tomatoes. Acknowledgements This study was supported by National Science and Technology support program (2006BAD10A04) the Teaching and Research Award Program for Outstanding Young Teachers in Higher Education Institutions of MOE, PRC, Natural Science Foundation of China (Project No: 30671213), Specialized Research Fund for the Doctoral
678
Y. Shao et al. / Journal of Food Engineering 81 (2007) 672–678
Program of Higher Education (Project No: 20040335034) and Natural Science Foundation of Zhejiang Province, China (Project No: RC02067). References Arazuri, S., Jaren, C., Arana, J. I., & Perez de Ciriza, J. J. (2007). Influence of mechanical harvest on the physical properties of processing tomato (Lycopersicon esculentum Mill.). Journal of Food Engineering, 80, 190–198. Artes, F., Conesa, M. A., Hernandez, S., & Gil, M. I. (1999). Keeping quality of fresh-cut tomato. Postharvest Biology and Technology, 17, 153–162. Batten, G. D. (1998). Plant analysis using near infrared reflectance spectroscopy: the potential and the limitations. Australian Journal of Experimental Agriculture, 38, 697–706. Cen, H. Y., He, Y., & Huang, M. (2006). Fast measurement of soluble solids contents and acidity in orange juice using chemometrics and Vis/ NIRS. Journal of Agricultural and Food Chemistry, 54, 7437–7443. Diezma-lglesias, B., Valero, C., Garcia-Ramos, F. J., & Ruiz-Altisent, M. (2006). Monitoring of firmness evolution of peaches during storage by combing acoustic and impact methods. Journal of Food Engineering, 77, 926–935. Gao, L., & Ren, S. X. (2005). Simultaneous spectrophotometric determination of four metals by two kinds of partial least squares methods. Spectrochimica Acta, Part A, 61, 3013–3019. Golic, M., & Walsh, K. B. (2006). Robustness of calibration models based on near infrared spectroscopy for the in-line grading of stonefruit for total soluble solids content. Analytica Chimica Acta, 555, 286–291. Gomez, A. H., He, Y., & Pereira, A. G. (2006). Non-destructive measurement of acidity, soluble solids and firmness of Satsuma mandarin using Vis/NIR-spectroscopy techniques. Journal of Food Engineering, 77, 313–319. Haalan, D., & Thomas, E. (1988). Partial Least Squares method for spectral analysis. 1. Relation to other quantitative calibration methods and the extraction of qualitative information. Analytical Chemistry, 60, 1193–1202. He, Y., Feng, S. J., Deng, X. F., & Li, X. L. (2006). Study on lossless discrimination of varieties of yogurt using the visible/NIR-spectroscopy. Food Research International, 39, 645–650. He, Y., Li, X. L., & Shao, Y. N. (2006). Discrimination of varieties of apple using near infrared spectra based on principal component analysis and artificial neural network model. Spectroscopy and Spectral Analysis, 26, 850–853. Jha, S., & Matsuoka, T. (2004). Non-destructive determination of acid– brix ratio of tomato juice using near infrared spectroscopy. International Journal of Food Science and Technology, 39, 425. Jong, S. (1993). PLS fits closer than PCR. Journal of Chemometrics, 7, 551–557. Kawano, S., Watanabe, H., & Iwamoto, M. (1992). Determination of sugar content in intact peaches by near infrared spectroscopy with fiber optics in interactance mode. Journal of Japanese Society for Horticultural Science, 61, 445–451. Lammertyn, J., Nicolai, B., Ooms, K., Smedt, V., & Baerdemaeker, J. (1998). Non-destructive measurement of acidity, soluble solids, and firmness of Jonagold apples using NIR-Spectroscopy. Transactions of the ASAE, 41, 1086–1094. Lu, R. (2001). Predicting firmness and sugar content of sweet cherries using near-infrared diffuse reflectance spectroscopy. Transactions of the ASAE, 44, 1265–1271.
Lu, R., Guyer, D., & Beaudry, R. (2000). Determination of firmness and sugar content of apple using NIR diffuse reflectance. Journal of Texture Studies, 31, 615–630. Martens, H., & Naes, T. (1987). Multivariate calibration by data compression. In P. Williams, K. Norris, & S. Paul (Eds.), Near-infrared Technology in the Agricultural and Food Industries (pp. 123–139). St. Paul, Minnesota: American Association of Cereal Chemists. McGlone, V. A., Fraser, D. G., Jordan, R. B., & Ku¨nnemeyer, R. (2003). Internal quality assessment of mandarin fruit by vis/NIR spectroscopy. Journal of Near Infrared Spectroscopy, 11, 323–332. McGlone, V. A., Jordan, R. B., & Martinsen, P. J. (2002). Vis/NIR estimation at harvest of pre- and post-storage quality indices for ‘Royal Gala’ apple. Postharvest Biology and Technology, 25, 135–144. McGlone, V. A., Jordan, R. B., Seelye, R., & Martinsen, P. J. (2002). Comparing density and NIR methods for measurement of Kiwifruit dry matter and soluble solids content. Postharvest Biology and Technology, 26, 191–198. McGlone, V., & Kawano, S. (1998). Firmness, dry-matter and solublesolids assessment of postharvest kiwifruit by NIR spectroscopy. Postharvest Biology and Technology, 13, 131–141. Mwithiga, G., Mukolwe, M. I., Shitanda, D., & Karanja, P. N. (2007). Evaluation of the effect of ripening on the sensory quality and properties of tamarillo (Cyphomandra betaceae) fruits. Journal of Food Engineering, 79, 117–123. Osborne, S., & Ku¨nnemeyer, R. (1999). A low-cost system for the grading of kiwifruit. Journal of Near Infrared Spectroscopy, 7, 9–15. Pedro, A. M. K., & Ferreira, M. M. C. (2005). Nondestructive determination of solids and carotenoids in tomato products by nearinfrared spectroscopy and multivariate calibration. Analytical Chemistry, 77, 2505–2511. Peiris, K. H. S., Dull, G. G., Leffler, R. G., & Kays, S. J. (1998). Nearinfrared spectrometric method for nondestructive determination of soluble solids content of peaches. Journal of the American Society for Horticulture Science, 123, 898–905. Peiris, K. H. S., Dull, G. G., Leffler, R. G., & Kays, S. J. (1998b). Nearinfrared spectrometric method for nondestructive determination of soluble solids content in processing tomatoes. Journal of the American Society for Horticulture Science, 123, 1089–1093. Peiris, K., Dull, G., Leffler, R., & Kays, S. (1997). Nondestructive determination of soluble solids content of peaches by near infrared spectroscopy. International Conference of Sensors for Nondestructive Testing: measuring the quality of fresh fruit and vegetables. NRAES: Ithaca, NY, pp.77–87. Peirs, A., Lammertyn, J., Ooms, K., & Nicolaı¨, B. M. (2000). Prediction of the optimal picking date of different apple cultivars by means of VIS: NIR-spectroscopy. Postharvest Biology and Technology, 21, 189–199. Saranwong, S., Sornsrivichai, J., & Kawano, S. (2004). Prediction of ripestage eating quality of mango fruit from its harvest quality measured nondestructively by near infrared spectroscopy. Postharvest Biology and Technology, 31, 137–145. Schmilovitch, Z., Mizrach, A., Hoffman, A., Egozi, H., & Fuchs, Y. (2000). Determination of mango physiological indices by nearinfrared spectrometry. Postharvest Biology and Technology, 19, 245–252. Slaughter, D. (1995). Nondestructive determination of internal quality in peaches and nectarines. Transactions of the ASAE, 38, 617–623. Slaughter, D., Barrett, D., & Boersig, M. (1996). Nondestructive determination of soluble solids in tomatoes using near infrared spectroscopy. Journal of Food Science, 61, 695–697. Wold, S., Esbensen, K., & Geladi, P. (1987). Principal component analysis. Chemometric and intelligent Laboratory System, 2, 37–52.