Prediction of chemical composition and geographical origin traceability of Chinese export tilapia fillets products by near infrared reflectance spectroscopy

LWT - Food Science and Technology 60 (2015) 1214e1218

Contents lists available at ScienceDirect

LWT - Food Science and Technology journal homepage: www.elsevier.com/locate/lwt

Prediction of chemical composition and geographical origin traceability of Chinese export tilapia fillets products by near infrared reflectance spectroscopy* Yuan Liu a, *, Dong-hong Ma a, Xi-chang Wang a, Li-ping Liu b, Yu-xia Fan a, Jin-xuan Cao c a b c

College of Food Science and Technology, Shanghai Ocean University, Shanghai 201306, PR China College of Fisheries and Life Science, Shanghai Ocean University, Shanghai 201306, PR China Life Science and Biotechnology College, Ningbo University, Ningbo, Zhejiang 315211, PR China

a r t i c l e i n f o

a b s t r a c t

Article history: Received 9 May 2013 Received in revised form 20 July 2014 Accepted 1 September 2014 Available online 16 September 2014

Near infrared reflectance spectroscopy (NIRS) analysis was used to predict proximate chemical composition of Chinese export tilapia fillets from four geographical origins (Guangdong Province, Hainan Province, Guangxi Province and Fujian Province, respectively). NIRS provided good reliability in the prediction of chemical composition of tilapia fillets but weak results in crude protein prediction. Origin traceability is an important part of food safety traceability system. The tilapia origin traceability model was developed by near infrared reflectance (NIR) spectroscopy coupled with soft independent modeling of class analogy (SIMCA). The result showed that when classifying tilapia by means of SIMCA, more than 80% of from the Guangdong, Hainan and Fujian systems and 75% of fillets from the Fujian system were correctly and exclusively assigned to the correctly and exclusively assigned to the corresponding clusters. No spectra were assigned to two or more clusters, while a certain number of spectra (10e18%) were not assigned to any class. Only 1e2% of samples were classified incorrectly. The results of this study indicated that NIRS coupled with pattern recognition methods was a feasible way for origin traceability of export tilapia fillets. © 2014 Elsevier Ltd. All rights reserved.

Keywords: Near infrared spectroscopy (NIRS) Origin traceability Soft independent modeling of class analogy (SIMCA) Chemical composition Tilapia

1. Introduction Tilapia fillet is one of important aquaculture fish products in the European Union. Trade in farmed tilapia from China to United States, Russia and the European Union is growing rapidly and becoming a significant component of Chinese export of aquatic products. Now, many fish productions are internationally traded with net flows from developing to developed countries. Growth in production and export of fish from Asia to European markets has accelerated over the last decade. The wide competition among producing countries in Asia and the consequent lowering of market prices are demanding the differentiation and characterization of tilapia fillet quality that has also occurred for other local foods. The

* *This manuscript was presented in the international conference of “Food Innova-2012” Hangzhou, China, December 12e14, 2012. * Corresponding author. Present address: Laboratory of Food Nutrition and Quality Evaluation, College of Food Science and Technology, Shanghai Ocean University, 999 Huchenghuan Road, Lingang New City, Pudong New District, Shanghai 201306, PR China. Tel.: þ86 21 61900380; fax: þ86 21 61900365. E-mail addresses: [email protected], [email protected] (Y. Liu).

http://dx.doi.org/10.1016/j.lwt.2014.09.009 0023-6438/© 2014 Elsevier Ltd. All rights reserved.

different rearing systems and geographical origin used for tilapia production may affect flesh quality, especially in terms of fat concentration and quality. Moreover, prices often differ widely according to origin and are highest for wild fish. Near infrared reflectance spectroscopy (NIRS) analysis may provide quick and wide information on food quality (Sun, Xu, & Ying, 2009), and has already been successfully used to predict the chemical composition of various fish and meats, such as salmon, ~ eda, & mackerel, halibut, capelin and beef (Alomar, Gallo, Castan Fuchslocher, 2003; Cozzolino, Murray, & Scaife, 2002; Nortvedt, Torrissen, & Tuene, 1998; Solberg & Fredriksen, 2001), but less information is available on tilapia. Accurate analysis of fish composition is very important because of its relationship to both the quality and specific characteristics, such as eating quality and impact on consumer health. Over the past five years, a lot of feasibility studies on the application of NIRS to measurement of chemical composition in meat, seafood and many other foods have been reported (Weeranantanaphan, Downey, Allen, & Da-Wen, 2011). In the NIRS analysis of fish or any other food products, sample preparation technique is one of the main problems due to the high absorbance of the NIRS signal by water, which masks and

Y. Liu et al. / LWT - Food Science and Technology 60 (2015) 1214e1218

disturbs information on the chemical constituents (Shenk, Workman, &Westerhaus, 1992). As described by Downey (1996a), appropriate statistical treatment of NIR spectra permitted to identify the origin and characterize of a wide range of vegetal foods (from fruit juice to coffee). Although NIRS has been successfully used in some studies on vegetal foods (Chen et al., 2008; Woo, Kim, Ze, & Chung, 2005) and ^ne, meat to identify species and some origins (Downey & Beauche 1997; McElhinney, Downey, & Fearn, 1999; Olivier, Sinnaeve, & Dardenne, 2000), it has been applied only rarely to fish geographical identification and characterization. The aim of this study is to propose the use of NIRS combined with SIMCA to create tilapia origin traceability model, to provide new idea for origin traceability of meat.

2. Material and methods 2.1. The set of sample and chemical analyses Over a 2-month period from October to December, 2010, tilapia fillets samples were obtained from four regions (Guangdong Province, Hainan Province, Guangxi Province and Fujian Province). The main characteristics of tilapia are summarized in Table 1. Only one commercial size of tilapia fillet, 5e7 ounces, was selected from each region. These differences in number were imposed by the product available at the regions. The tilapia fillets were used for NIRS and chemical analyses. The fillets were analyzed for proximate composition following Chinese standard methods. Fat, protein and moisture contents in tilapia were in accordance with GB/T 9695.1-2008/ISO 1444: 1996 (Meat and meat products e Determination of free fat content), GB/T 9695.11-2008 (Meat and meat products e Determination of nitrogen content) and GB/T 9695.15-2008 (Meat and meat products e Determination of moisture content) respectively.

2.2. NIRS analysis NIRS analysis was performed by a near infrared reflectance spectroscopy spectrometer (NIRFlex N-500, Büchi Labortechnik AG, Flawil, Switzerland) in the 4000 and 10,000 cm1 range with a 4 cm1 step. The number of scans was 64. Each fillet was analyzed according to the following sample: dorsal flesh. Each dorsal flesh sample was treated with meat grinder. Then the treated fish meat paste was put into specific petri dish. Scanning temperature of NIR was controlled in the 20 and 25  C ranges. Calibration equations were calculated by partial least square regression (PLSR) to predict moisture (g/100 g), crude protein (g/100 g), and lipid (g/100 g) content. The number of factors used as independent variables in the prediction equations was fixed at a maximum of 20 in order to avoid over fitting (Shenk & Westerhaus, 1994). The optimal number of factors was chosen as a function of the first local minimum in the validation residual variance plot. Full cross-validation was used. Prediction equations were evaluated in terms of coefficient of determination in calibration (R2c) and cross-validation (R2cv),

Table 1 The geographical origin and feeding mode of tilapia. Region

Longitude

Latitude

Sampling time

Number

Guangdong Hainan Guangxi Fujian

110 310 110 100 108 030 119 400

20 560 19 580 22 080 27 280

2010.10e2010.11 2010.11e2010.12 2010.12 2010.12

58 56 48 46

1215

standard error of calibration (SEC) and standard error of crossvalidation (SECV). Principal component analysis (PCA) was used to calculate models for four clusters of samples on the basis of the geographical system using full cross-validation. Soft Independent Modeling Class Analogy (SIMCA) was used to measure the model-to-model distance and classify samples by the geographical origin. Cluster membership in SIMCA was tested at P < 0.05 level. Both PCA and SIMCA were performed using BUCHINIR Cal 5.2 software.

3. Results 3.1. NIRS prediction of chemical composition The proximate composition of the tilapia fillets were analyzed separately and averaged to obtain a reference value. Both moisture (60.9e80.2 g/100 g) and lipid (1.1e14.4 g/100 g) concentrations varied widely, while crude protein varied in a closer range (16.1e22.8 g/100 g). Table 2 showed the average and SD value of different sample sets. Calibration and validation results for the prediction of chemical composition of tilapia fillets are reported in Table 3. Predictions of moisture and lipid calculated on the spectra of minced fillets showed high correlation both in calibration and cross-validation, while prediction of protein showed low correlation. NIRS prediction of moisture concentration provided good coefficients of determination using dorsal flesh of tilapia samples in Calibration set and validation set (R2c ¼ 0.95 and R2cv ¼ 0.95) with a very low prediction error (SEC ¼ 0.85 g/100 g and SECV ¼ 0.87 g/ 100 g). The same accuracy was observed for lipid prediction in calibration set and validation set (R2c ¼ 0.97, R2cv ¼ 0.97, SEC ¼ 0.68 g/100 g and SECV ¼ 0.66 g/100 g). NIRS prediction of crude protein in tilapia fillets was less successful. Coefficients of determination gave fairly good results in Calibration set and validation set (R2c ¼ 0.61, SEC ¼ 0.39 g/100 g, R2cv ¼ 0.30, SECV ¼ 0.55 g/100 g). The comparison of the chemical values and NIR predicted values of moisture and lipid were shown in Figs. 1 and 2. The abscissa was the measured value by chemical methods; and the ordinate was the calculated values by NIRS. From the two figures, we could find that all plot points were uniformly distributed in the line around, so the moisture and lipid prediction models were effective, that is, model forecasting capability is very good. Randomly selected some model-outside samples were used for external validation of the model. The comparison of external validation results was shown in Tables 4 and 5. The largest relative

Table 2 Chemical composition of tilapia fillets. Moisture (g/100 g) Guangdong (no. ¼ 66) Average 76.1 SD 1.5 Hainan (no. ¼ 65) Average 70.7 SD 3.6 Guangxi (no. ¼ 50) Average 67.6 SD 2.3 Fujian (no. ¼ 49) Average 70.3 SD 1.7 All samples (no. ¼ 208) Average 70.9 SD 3.8

Crude protein (g/100 g)

Lipid (g/100 g)

19.1 0.6

2.7 1.5

19.1 0.7

8.5 3.7

19.3 0.7

10.8 2.8

19.3 0.6

8.3 2.1

19.2 0.7

7.9 3.9

1216

Y. Liu et al. / LWT - Food Science and Technology 60 (2015) 1214e1218

Table 3 Coefficients of determination in calibration (R2c) and cross-validation (R2cv) and standard errors of calibration (SEC) and cross-validation (SECV) obtained by partial least-square regression to predict the chemical composition of tilapia fillets analyzed by NIRS. Calibration

Moisture (g/100 g) Crude protein (g/100 g) Lipid (g/100 g)

Full crossvalidation

Factors

R2c

SEC

R2cv

SECV

3 7 2

0.952 0.613 0.977

0.852 0.391 0.685

0.959 0.306 0.971

0.872 0.550 0.706

error between calculated value by moisture NIR model and the measured value by chemical methods was only 1.67 g/100 g. And the largest relative error between calculated value by lipid NIR model and the measured value by chemical methods was only 6.48 g/100 g. So the established NIR model could well predict the contents of moisture and lipid of tilapia products.

3.2. SIMCA analysis of spectral data SIMCA analysis was performed on spectra taken from minced tilapia fillets. As a first step, within cluster principal component analysis selected 5, 6, 7 and 5 principal components for Guangdong, Hainan, Guangxi, and Fujian province, respectively. The highest model-to-model distance (>10) was measured between the

Table 4 External validation results of models for moisture(g/100 g). Moisture (Exp.) Moisture (Cal.) Absolute error (Cal.  Exp.) Relative error (%) 73.01 71.87 70.92 71.41 77.05 73.15 69.85 75.01

73.08 71.76 70.91 72.22 76.48 74.46 69.22 75.08

0.07 1.11 0.01 0.81 0.57 1.31 0.63 0.07

0.09 1.42 0.02 1.04 0.72 1.67 0.79 0.09

Guangdong system cluster and Fujian system clusters. The lowest distance was observed between Guangdong and Hainan systems. In SIMCA, the model-to-model distance was higher or lower to other three (the minimum distance required for a suitable cluster separation) when comparing the Guangdong with others (from 3.90 with respect to the Hainan to 13.59 with respect to the Fujian) (Table 6). When classifying tilapia by means of SIMCA, more than 80% of samples from the Guangdong, Hainan and Fujian systems and 75% of fillets from the Guangxi system were correctly and exclusively assigned to the correctly and exclusively assigned to the corresponding clusters (Table 7). No spectra were assigned to two or more clusters, while a certain number of spectra (10e18%) were not assigned to any class. The percentage of wrong classification was very low (1e2%).

Fig. 1. Scatter plots of the chemical values and NIR predicted values for moisture in tilapia products.

Fig. 2. Scatter plots of the chemical values and NIR predicted values for lipid in tilapia products.

Y. Liu et al. / LWT - Food Science and Technology 60 (2015) 1214e1218 Table 5 External validation results of models for lipid (g/100 g).

Table 7 Performance of SIMCA models of tilapia from four regions.

Lipid (Exp.)

Lipid (Cal.)

Absolute error (Cal.  Exp.)

Relative error (%)

1.12 3.53 2.03 3.71 3.64 3.71 2.64

1.46 4.10 1.84 2.82 2.78 4.54 2.45

0.34 0.57 0.19 0.89 0.86 0.83 0.19

3.07 4.20 1.61 6.48 6.33 6.04 1.48

4. Discussion Protein, moisture and fat are important functions and nutritional components of fish and fish products. It is often difficult to compare the reported results for measurement given that they are not the same species, size, portions and so on. In recent years, more and more applications of NIRS analysis in vegetable foods have been performed with special attention on minimal sample preparation, but seldom in meats. So in our study, the well minced tilapia fillets were analyzed in the 4000 and 10,000 cm-1 range respectively, and obtained satisfactory results. Early periods, Downey (1996b) attempted to analyze skinned farmed salmon by a fiber optic probe in the 400e1100 nm range and obtained satisfactory results for moisture (R2c ¼ 0.69; SEP ¼ 1.45 g/100 g) and lipid (R2c ¼ 0:70; SEP ¼ 2.04 g/100 g). Better results were achieved for lipid prediction (R2c ¼ 0:79; SEP ¼ 1.1 g/100 g) by Solberg (2000) on salmons analyzed. Moreover, NIRS accuracy improved greatly when analyzing the salmon fillet (R2c ¼ 0:88; SEP ¼ 0.82 g/100 g). Ortiz-Somovilla (Ortiz~ a-Espan ~ a, Gait rez-Aparicio, & De Somovilla, Espan an-Jurado, Pe Pedro-Sanz, 2007) got good results of homogenized and minced mass of pork sausages by NIRS for fat, moisture and protein respectively. Minced sausage major fat, moisture and protein were t: 0.98, 0.98 and 0.93 (R2) and 1.38 g/100 g, 1 g/100 g, 0.83 g/100 g (SEP); homogenized sausage meats were 0.99, 0.98 and 0.93 (R2), and 0.94 g/100 g, 0.76 g/100 g and 0.87 g/100 g (SEP), respectively. Better results could have been achieved by correlating the spectra of each portion with its exact composition, as performed by Lee (Lee, Cavinato, Mayes, & Rasco, 1992), rather than with the composition of the whole fillet. In fact, the heterogeneity of fat distribution along the fillet and between the dorsal and ventral sites has been proved (Downey, 1996b; Fjellanger, Obach, & Roselund, 2001). So in our study, we used portions of tilapia for study. Compared to our results on pretreated fillet portions, we provided more reliable calibrations for chemical composition. Calibrations for the prediction of moisture and lipid concentrations obtained in our study analyzing portions of tilapia fillets were quite satisfactory. In our study, prediction accuracy for moisture, lipid increased remarkably with tilapia fillets but remained only fair for protein concentration. Isaksson (Isaksson, Tøgersen, Iversen, & Hildrum, 1995), analyzing ground salmon fillets, obtained higher prediction accuracy for protein (SECV ¼ 0.37 g/100 g), while fat and

Table 6 Mahalanobis distances between classes in Soft Independent Modeling of Class Analogy (SIMCA) of tilapia fillets from four regions.

Guangdong Hainan Guangxi Fujian

1217

Guangdong

Hainan

Guangxi

Fujian

1 e e e

3.90 1 e e

9.30 7.04 1 e

13.59 10.07 11.53 1

Origin

Correctly classified (%)

Classified in two or more clusters (%)

Wrongly classified (%)

Not classified (%)

Guangdong Hainan Guangxi Fujian

85 82 75 83

0 0 0 0

0 2 2 1

11 12 18 10

moisture prediction was similar to our results. However, Cozzolino (Cozzolino et al., 2002) successfully predicted total volatile nitrogen (R2v ¼ 0.83; SECV ¼ 0.35 g/100 g) in fresh minced raw fish (various species). Good prediction accuracy by NIRS for chemical composition was confirmed by Solberg and Fredriksen (2001) in fresh minced capelin at different physiological states. The lower prediction performance that we obtained compared to other studies most likely depended on the wide variations of chemical characteristics in our set of tilapia fillets, which included great differences in geographical origin and fish rearing systems. A more specific and local calibration would have probably provided more accurate but less robust prediction equations. Liu (Liu, Li, Liu, Huang, & Ji-cheng, 2010) successfully predicted the origin of dry red grape wine from Changli, Yantai and Shacheng. The results showed that the accuracy rates were 90％and 86.7％ respectively by using the original sample and the remaining one cross-validation, which could meet the requirements of qualitative discrimination. This result is better than ours, maybe because orange juice is a liquid, which is better uniformity of the sample for detection. So in future study, we maybe should devote ourselves to sample treatments. Besides, in the meat area, Zhang (2008) succeed in tracing the origin of mutton from Jining city of Shandong Province, Hebei Province, Linhe city of Inner Mongolia and Rinchuan city of Ningxia region, respectively. It indicates that the technique of NIR was a feasible way for tracing the origin of meats. Xiccato (Xiccato, Trocino, Tulli, & Tibaldi, 2004) succeed in forecasting chemical composition and geographical origin of European sea bass (Dicentrarchuslabrax L.) by NIRS and SIMCA. In our study, some pretreatments of tilapia fillets were necessary to obtain satisfactory classification of tilapia by origin. Spectral variations among clusters did not depend exclusively on the geographical origin but more probably on a combination of other factors (feeding regime, growth pattern, muscular activity, competition, moisture quality, etc.).

5. Conclusions NIRS showed good reliability in the prediction of moisture and lipid, but weaker results for crude protein in tilapia fillets characterized by wide variation in chemical composition. NIRS analysis succeeded in giving reliable information on the tilapia fillets origin system, which the homogeneity of the samples was required for correct classification by origin. Our results indicate NIRS as a promising method for tilapia fillets characterization, traceability and authentication.

Conflict of interest statement We declare that we have no financial and personal relationships with other people or organizations that can inappropriately influence our work; there is no professional or other personal interest of any nature or kind in any product, service and/or company that

1218

Y. Liu et al. / LWT - Food Science and Technology 60 (2015) 1214e1218

could be construed as influencing the position presented in, or the review of, the manuscript entitled. Acknowledgments This work was supported by the EU 7th frame project “Sustaining Ethical Aquaculture Trade” (Grant agreement no. 222889) and National Natural Science Foundation of China (Grant No. 30901125). References ~ eda, M., & Fuchslocher, R. (2003). Chemical and Alomar, D., Gallo, C., Castan discriminant analysis of bovine meat by near infrared reflectance spectroscopy (NIRS). Meat Science, 63(4), 441e450. Chen, Y., Xie, M.-Y., Yan, Y., Zhu, S.-B., Nie, S.-P., Li, C., et al. (2008). Discrimination of Ganoderma lucidum according to geographical origin with near infrared diffuse reflectance spectroscopy and pattern recognition techniques. Analytica Chimica Acta, 618(2), 121e130. Cozzolino, D., Murray, I., & Scaife, J. R. (2002). Near infrared reflectance spectroscopy in the prediction of chemical characteristics of minced raw fish. Aquaculture Nutrition, 8(1), 1e6. Downey, G. (1996a). Authentication of food and food ingredients by near infrared spectroscopy. Journal of Near Infrared Spectroscopy, 4, 47e62. Downey, G. (1996b). Non-invasive and non-destructive percutaneous analysis of farmed salmon flesh by near infra-red spectroscopy. Food Chemistry, 55(3), 305e311. ^ne, D. (1997). Discrimination between fresh and frozenDowney, G., & Beauche then-thawed beef m. longissimus dorsi by combined visible-near infrared reflectance spectroscopy: a feasibility study. Meat Science, 45(3), 353e363. Fjellanger, K., Obach, A., & Roselund, G. (2001). Proximate analysis of fish with special emphasis on fat. Oxford, UK: Fishing News Books, Blackwell. GB/T 9695.1-2008. (2008). Meat and meat products e Determination of free fat content. China. GB/T 9695.11-2008. (2008). Meat and meat products e Determination of nitrogen content. China. GB/T 9695.15-2008. (2008). Meat and meat products e Determination of moisture content. China. Isaksson, T., Tøgersen, G., Iversen, A., & Hildrum, K. I. (1995). Non-destructive determination of fat, moisture and protein in salmon fillets by use of nearinfrared diffuse spectroscopy. Journal of the Science of Food and Agriculture, 69(1), 95e100. Lee, M., Cavinato, A., Mayes, D., & Rasco, B. (1992). Noninvasive short-wavelength near-infrared spectroscopic method to estimate the crude lipid content in the

muscle of intact rainbow trout. Journal of agricultural and food chemistry, 40(11), 2176e2181. Liu, Wei, Li, De-mei, Liu, Guo-jie, Huang, Wei-dong, & Ji-cheng, Z. (2010). Application of near infrared spectroscopy and Fisher discriminant function to distinguish the original production place of grape wine. LiquoreMaking Science & Technology, 7, 65e69. McElhinney, J., Downey, G., & Fearn, T. (1999). Chemometric processing of visible and near infrared reflectance spectra for species identification in selected raw homogenised meats. Journal Near Infrared Spectroscopy, 7(3), 145e154. Nortvedt, R., Torrissen, O. J., & Tuene, S. (1998). Application of near-infrared transmittance spectroscopy in the determination of fat, protein and dry matter in Atlantic halibut fillet. Chemometrics and intelligent laboratory systems, 42(1), 199e207. re, Sinnaeve, G., & Dardenne, P. (2000). Attempted authentication of Olivier, Fumie cut pieces of chicken meat from certified production using near infrared spectroscopy. Journal of Near Infrared Spectroscopy, 8(1), 27e34. ~ a-Espan ~ a, F., Gaita n-Jurado, A., Pe rez-Aparicio, J., & De Ortiz-Somovilla, V., Espan Pedro-Sanz, E. (2007). Proximate analysis of homogenized and minced mass of pork sausages by NIRS. Food Chemistry, 101(3), 1031e1040. Shenk, J. S., & Westerhaus, M. O. (1994). The application of near infrared reflectance spectroscopy (NIRS) to forage analysis. Lincoln, Nebrasks, USA: Crop Science Society of America. Shenk, J. S., Workman, J. J., & Westerhaus, M. O. (1992). Application of NIRS spectroscopy to agricultural products. New York, USA: Marcel Dekker. Solberg, C. (2000). Diode-array near infrared analysis of whole and filleted salmon. West Sussex, UK: NIR. Solberg, C., & Fredriksen, G. (2001). Analysis of fat and dry matter in capelin by near infrared transmission spectroscopy. Journal of Near Infrared Spectroscopy, 9(3), 221e228. Sun, T., Xu, H. R., & Ying, Y. B. (2009). Progress in application of near infrared spectroscopy to nondestructive on-line detection of products/food quality. Guang pu xue yu guang pu fen xi¼ Guang pu, 29(1), 122. Weeranantanaphan, J., Downey, G., Allen, P., & Da-Wen, S. (2011). A review of near infrared spectroscopy in muscle food analysis: 2005-2010. Journal of Near Infrared Spectroscopy, 19(2), 61e104. Woo, Y. A., Kim, H. J., Ze, K. R., & Chung, H. (2005). Near-infrared (NIR) spectroscopy for the non-destructive and fast determination of geographical origin of Angelicae gigantis Radix. Journal of pharmaceutical and biomedical analysis, 36(5), 955e959. Xiccato, G., Trocino, A., Tulli, F., & Tibaldi, E. (2004). Prediction of chemical composition and origin identification of european sea bass (Dicentrarchus labrax L.) by near infrared reflectance spectroscopy (NIRS). Food Chemistry, 86(2), 275e281. Zhang, N., Zhang, D. Q., Li, S. R., & Li, Q. P. (2008). Preliminary study on origin traceability of mutton by near infrared reflectance spectroscopy coupled with SIMCA method. Transactions of the Chinese Society of Agricultural Engineering, 12.