Opt Quant Electron (2017)49:1 DOI 10.1007/s11082-016-0848-8
Terahertz spectroscopy and chemometric tools for rapid identification of adulterated dairy product Jianjun Liu1
Received: 27 June 2016 / Accepted: 30 November 2016 Springer Science+Business Media New York 2016
Abstract Terahertz spectroscopy has been investigated as a quick and non-destructive evaluation method to identify adulterated dairy products. Compared with the traditional identification methods, terahertz spectroscopy measurements can easily distinguish different adulterated dairy products according to the fatty acids, without pretreatment of the sample. Terahertz spectra are collected from samples of whole samples both without pretreatment. The difference between the spectra of samples can be observed with the fatty acids. This paper is using terahertz spectroscopy combination with chemometric tools to identify samples with PCA and SVM-DA. All samples can be correctly identified by SVMDA models. These results demonstrate the performance of terahertz spectroscopy couple with chemometrics methods to identify adulterated food. Keywords Terahertz Spectroscopy Chemometrics Food Fatty acids
1 Introduction Adulterated dairy products, especially the milk, are subject of food safety research, which followed by research reports, peer review and so on (Spink and Moyer 2011). Adulterated milk with glucose not only is a sneaky behavior, but also an illegal. While how to fast and efficiently detect adulterated dairy products, there is seldom resolved in the literature. In analytical chemical area, chemometric methods are considered as an effectively tool to identify and control adulterated dairy products. Souza et al. using hierarchical cluster analysis (HCA) and principal component analysis (PCA) to verify adulterated milk in different manufactory. The chemometric methods are able to show the ability of analysing data is mainly based on pattern recognition, classical technique and discriminant analysis.
& Jianjun Liu
[email protected] 1
School of Electrical Engineering, Jiujiang University, Jiujiang 332005, Jiangxi, China
123
1 Page 2 of 8
J. Liu
Barker and Rayens (2003) proposed a novel classification technique, namely PLS-DA, which combine partial least square with discriminant analysis. However, the object of above chemometric methods is physicochemical data of sample depicted by IOC regulation (Aquino et al. 2014; Cruz et al. 2013; Matera et al. 2014; Zielinski et al. 2014). With the development of spectral technology, several applications of spectroscopy technology, such as Raman and NIR (near infrared), are used to detect dairy products (Cheng et al. 2010; Moros et al. 2007), but there are rarely reports on identification of adulterated dairy products using terahertz spectroscopy. Terahertz is an electromagnetic wave, whose frequency range is 0.1–10 THz (wavelength 30–3; Liu et al. 2015a, b, 2016a, b; Xi-Liang et al. 2013; Liu and Li 2014). Studies have shown that the vibration and rotational energy levels of most biological molecules locate in the THz band. Duo to the terahertz have potential application of security, biological, medical detection and so on. Based on above, this paper proposed a rapid and alternative method for the detection of fat powder in dairy products and the classification of the category of dairy products (Highfat and Low-fat) by employing terahertz spectroscopy and chemometric methods. Support Vector Machine-discriminate analysis (SVM-DA) is applied to identify adulterated and non-adulterated dairy products.
2 Materials and methods 2.1 Samples This paper takes milk powder as the object of the research, those samples are divided into four groups, skim milk (20 samples), low fat milk (20 samples), skim milk adulterated with fat powder (20 samples) and low fat milk adulterated with fat powder (20 samples), a total of 80 samples. The sample of skim milk supplied by Rain Bow supermarket; The low fat samples are obtained by adding the lipase enzyme to skim milk with concentration of 0.5 g/L, and the hydrolysis time is 5 h. All samples are desiccated in Labplant Spray Dryer SD-Basic (YITAI Weaving Machine Co., Ltd). The adulterated samples, a mixture of fat powder and milk are performed with varying concentration between 2.5 and 50% (w/w). The Samples were stored and packed at room temperature and avoid light.
2.2 Terahertz measurements Terahertz spectra of all samples are with a terahertz time-domain spectrometer. Each spectrum is an average of 30 scans collected from the spectrometer (Zomega Terahertz Corp., USA), where the center wavelength as the laser is 780 nm. To ensure the accuracy of the experiment, in the system, dry air is injected until the internal relative humidity to 2% below. Indoor relative humidity is 25%, and the temperature is 292 K. The standard spectrum is gathered before measure each sample. The Teralyzer software program is used for terahertz data acquisition.
2.3 Chemometric analysis All of the chemometrics methods are accomplished by using TQ Analyst V8.0 (Thermo Nicolet Corporation, Madison, WI, USA). THz spectra data of samples are collected from
123
Terahertz spectroscopy and chemometric tools for rapid…
Page 3 of 8 1
0.1 to 1.5 THz in this paper. Each of samples is composed a spectral matrix whose size is 256 9 512. Terahertz spectra are employed in data preprocessing by of Savitzky–Golay derivative with the first derivative. In order to reduce the influence of the external environment, terahertz spectra data of all samples are normalized. After preprocessing, Terahertz data preprocessing of samples is carried out by PCA which can preliminary evaluate the performance of discriminant of terahertz spectra in the region of 0.1–1.5 THz. SVM is a learning machine that based on the theory of Vapnik–Chervonenkis (VC) and structural risk minimization (SRM). By seeking the minimum risk of structured to improve the generalization ability of learning machine, achieving minimize of empirical risk and confidence limit, achieved good statistical under the condition of Less sample. The purpose of SVM is achieving the best generalization ability. SVM-DA is employed to regression and the classification of multiple classes. For SVM-DA, the data of samples are divided into two datasets: training set (60 samples) and evaluation set (20 samples). For classify samples as adulterated and unadulterated samples. The model of SVM-DA will create a dummy matrix which using 1 express adulterated sample and 0 express non-adulterated ones. Similarly, for classifying skim and low-fat samples, SVM-DA will create an another dummy matrix containing the 1 for the skim milk and 1 for the low-fat samples. The correct classification, sensitivity and specificity are used to evaluate the performance of the SVM-DA model.
3 Results and discussion 3.1 Characterization of samples using terahertz spectra The terahertz absorption spectra of skim milk and low-fat milk are shown in Fig. 1, and the main absorption frequency of different functional groups are reported by Almeida and Gelder (Almeida et al. 2011; Gelder et al. 2007). It can be seen that the most useful absorption frequency for this purpose is those at 0.384, 1.16, 1.196, 1.343 and 1.422 THz, which were assigned to the presence of fatty acids in the skim milk. For low-fat samples, in addition to the intensity decrease of the absorption frequency associated with the presence of fatty acids, it is found the presence of absorption frequency at 0.384, 1.16, 1.196, 1.346 and 1.434 THz, where principally absorption frequency 1.16 and 1.346 THz are concerned with the presence of CH2.
3.2 Vibrational spectra of the adulterated samples In order to more clearly understand the sensitive response of different samples, we select all samples to do 30 random tests, the purpose of this test is to demonstrate the error range of system, we select one experimental data as the reference data from 30 groups of experimental data, the error analysis is carried out on other data. The terahertz spectra of pure fat powder, pure skim milk, and skim milk with fat powder added (10, 30, and 50% w/w) are shown in Fig. 2. Fat powder is a result of fatty acid calcium hydrolysis, and so its spectrum is related to the spectra of the molecules of carboxyl and hydrocarbon chain. The mainly absorption frequency of this adulterated agent at 0.096, 1.196, 1.16, 1.343 and 1.422 THz which respectively corresponding to the deformation of endocyclic and exocyclic.
123
1 Page 4 of 8
J. Liu
Fig. 1 Terahertz absorbance of skim and low-fat milk
Fig. 2 Terahertz absorbance spectroscopy of samples
THz absorption characteristics in 1.2 and 0.348 THz is relative to C=O and C=C stretching and C–O–C deformation modes; the absorption peaks of 0.348 and 1.16 THz are also consider as the fingerprint, and is frequently cited in literature by other scholars (Benzerdjeb et al. 2007; Nikonenko et al. 2005; Yang and Zhang 2009; Baranska et al.
123
Terahertz spectroscopy and chemometric tools for rapid…
Page 5 of 8 1
2005). Vibrations in the region of 1.1–1.4 THz are generally caused by C–C=O and C–C– C deformations. In this region, we can observe a very strong coupling which is related to the hydrocarbon skeletal deformations. The intense absorption characteristics at 0.38–1.5 THz have been used as a feature information to identify the presence of fat in different samples, as well as to characterize the carboxyl and hydrocarbon chain which is the constituents or derivatives of fat, such as fatty acid calcium. Adjust the relative concentration of milk and fat powder, the absorption spectra of the sample are indicative of changing lipid content and type concentration, which may be consociated with the presence of fat powder. By comparing the terahertz spectra of the pure skim milk with the sample containing 10% (w/w) of added fat powder, it is possible to identify tiny changes between adulterated and non-adulterated samples using the terahertz absorption spectrum which related to the presence of fat powder increase. Furthermore, modify the content of fat powder, the terahertz absorption spectrum will change too, as is observed for the sample containing 30% (w/w) of fat powder or more. It can be seen from Fig. 2 that the concentration of at least 30% (w/w) fat powder in milk will change the terahertz spectrum profile that is mainly characteristic of fat powder. Therefore, it is possible to rapidly identify the presence of fat powder without any sample preparation, which is very useful for field analysis.
3.3 PCA of the terahertz spectra data of samples The principal component analysis (PCA) is used to extract feature information of sugarcane samples. PCA indicate that the first two eigenvector apprehend more than 91.99% of the total variance. Figure 3 gives the score value of 80 samples by using PCA. It can be observed from Fig. 3 that most samples are divided into two regions, where region one involves most of the skim milk samples (including skim milk and skim milk with fat powder) and region two contains most of low-fat milk samples (including low-fat milk and
Fig. 3 Score plots of PC1 (76.63% variance) and PC2 (15.36% variance) for samples
123
1 Page 6 of 8
J. Liu
low-fat milk with fat powder). It is also observed a dispersion of the adulterated samples with higher concentrations of fat powder away from the non-adulterated samples. However, when the concentration of fat powder in milk below 30%, the samples are hard to separate accurately, so then requiring a chemometric method for better identification of the samples. The mainly responsible for samples differentiation could be detected by the variables absorption for each variable of the first principal component. It is noteworthy that the absorption frequency at 0.745 and 1.262 THz, the presence of fat powder in milk samples (skim milk and low-fat milk), had the highest information in PC1 and fingerprint characteristics of their own for group discrimination. According to the information of PC2, the absorption frequency at 0.661 THz, related to the presence of fat, are main factors for the separation of samples with fat, while the absorption at 0.745 and 1.262 THz related to the presence of fat acid, are responsible for arranging the low-fat samples.
3.4 Chemometric classification of adulterated samples In this paper, PCA and SVM-DA are used to classify different sample. SVM-DA is used to obtain a better classification of the terahertz spectra of milk samples without the addition of fat powder from adulterated ones. For all samples (30 non-adulterated samples and 30 adulterated samples, with the concentration of fat powder in milk from 10 to 50%), the SVM-DA model use the full terahertz range (0.1–1.5 THz). The resulting dataset are randomly divided into training set (20 non-adulterated samples and 20 adulterated samples) test set (10 non-adulterated and 10 adulterated samples) by cross-validation [using six latent variables (LV)]. The performance of SVM-DA model is shown in Fig. 4. In SVMDA model, 1 expresses adulterated sample and 0 express non-adulterated samples. It can be seen from Fig. 4 that the experiment measured value is close to 0 or 1. Under the condition of classification rate of 100% of adulterated and non-adulterated samples in the training set, the specificity is 100% and the sensitivity is 88.62%. For test set, the classification effects of SVM-DA are satisfactory. The experimental results show that the proposed method has advantages of good reliability and robustness.
3.5 Chemometric classification of low-fat samples The resulting dataset are randomly divided by cross-validation [using three latent variables (LVs)], completely different from other previously employed. Figure 5 shows the classification results of SVM-DA model. Under the condition of the classification rate of 100% of skim milk and low-fat milk samples in the training set, the specificity is 100% and the sensitivity is 98.62%. The results showed that proposed can contribute to the identification of low-fat dairy products. Terahertz spectroscopy combine with chemometrics method is not only able to classify The same kind of milk sample but also to produce an absorption fingerprint of different kinds of milk samples as showed for fat powder addition and for lowfat milk.
4 Conclusions Terahertz spectroscopy combine with chemometrics methods is a powerful tool for classification of adulterated and non-adulterated dairy products and can provide the advantage of escaping time-consuming, costly chemical and sensory analyses. Distinguishing
123
Terahertz spectroscopy and chemometric tools for rapid…
Page 7 of 8 1
Fig. 4 Results of the training set of SVM-DA model
Fig. 5 Results of the training set of the SVM-DA
adulterated products using this technique is meritorious, and this study exhibits the potential of terahertz spectroscopy couple with chemometrics methods for dairy products. The aim of further researches is to establish more worthful and powerful identification models for other products.
123
1 Page 8 of 8
J. Liu
Acknowledgements This work is supported by Guangxi Key Laboratory of Automatic Detecting Technology and Instruments (No. YQ16204).
References Almeida, M.R., Oliveira, K.S., Stephani, R., Oliveira, L.F.C.: Fouriertransform Raman analysis of milk powder: a potential method for rapid quality screening. J. Raman Spectrosc. 42, 1548–1552 (2011) Aquino, L.F.M.C., Silva, A.C.O., Freitas, M.Q., Felicio, T.L., Cruz, A.G., Conte Junior, C.A.: Identifying cheese whey an adulterant in milk: limited contribution of a sensometric approach. Food Res. Int. 62, 233–237 (2014) Baranska, M., Schulz, H., Baranski, R., Nothnagel, T., Christensen, L.P.: In Situ simultaneous analydes of polyacetylenes, carotenoids and polysaccharide in carrot roots. J. Agric. Food Chem. 53, 6565–6571 (2005) Barker, M., Rayens, W.: Partial least squares for discrimination. J. Chemom. 17, 166–173 (2003) Benzerdjeb, A.M., Mokhtari, I.N.T., Rahal, M.S.: Normal coordinates analyses of disaccharides constituted by D-glucose, D-galactose and D-fructose units. Spectrochim. Acta Part A 68, 284–299 (2007) Cheng, Y., Dong, Y., Wu, J., Yang, Y., Bai, H., Zheng, H., Li, M.: Screening melamine adulterant in milk powder with laser Raman spectrometry. J. Food Compost. Anal. 23, 199–202 (2010) Cruz, A.G., Cadena, R.S., Alvaro, M.B.V.B., Sant’Ana, A.S., Oliveira, C.A.F., Faria, J.A.F., Bolini, H.M.A., Ferreira, M.M.C.: Assessing the use of different chemometric techniques to discriminate low-fat and full-fat yogurts. LWT Food Sci. Technol. 50, 210–214 (2013) Gelder, J., Gussem, K., Vandenabeele, P., Moens, L.: Reference database of Raman spectra of biological molecules. J. Raman Spectrosc. 38, 1133–1147 (2007) Liu, J., Li, Z.: The terahertz spectrum detection of transgenic food. Optik 125(23), 6867–6869 (2014) Liu, J., Li, Z., Hu, F., Chen, T., Du, Y., Xin, H.: Identification of transgenic organisms based on terahertz spectroscopy and hyper sausage neuron. J. Appl. Spectrosc. 82(1), 104–110 (2015a) Liu, J., Li, Z., Hu, F., Chen, T., Du, Y., Xin, Haitao: Hyper sausage neuron: recognition of transgenic sugarbeet based on terahertz spectroscopy. Opt. Spectrosc. 118(1), 175–180 (2015b) Liu, J., Fan, L., Ku, J., Mao, L.: Absorber: a novel terahertz sensor in the application of substance identification. Opt. Quant. Electron. 48(2), 80–87 (2016a) Liu, J., Mao, L., Ku, J., He, J.: THz spectroscopy detection method for GMOs based on adaptive particle swarm optimization. Opt. Quant. Electron. 48(2), 167–173 (2016b) Matera, J.A., Cruz, A.G., Raices, R.S.L., Silva, M.C., Nogueira, L.C., Quite´rio, S.L., Ju´nior, C.C.: Discrimination of Brazilian artisanal and inspected pork sausages: application of unsupervised, linear and non-linear supervised chemometric methods. Food Res. Int. 64, 380–386 (2014) Moros, J., Garrigues, S., Guardia, M.: Evaluation of nutritional parameters in infant formulas and powdered milk by Raman spectroscopy. Anal. Chim. Acta 593, 30–38 (2007) Nikonenko, N.A., Buslov, D.K., Sushko, N.I., Zhbankov, R.G.: Spectroscopic manifestation of stretching vibrations of glycosidic linkage in polysaccharides. J. Mol. Struct. 752, 20–24 (2005) Spink, J., Moyer, D.C.: Defining the public health threat of food fraud. J. Food Sci. 76, 157–163 (2011) Xi-Liang, C., Xin, C., Zhi-Yong, Z.: THz-TDS spectra study of polymer materials with different polarity. J. Infrared Mullimeter Waves 32, 150–155 (2013) Yang, L., Zhang, L.M.: Chemical structural and chain conformational characterization of some bioactive polysaccharides isolated from natural sources. Carbohydr. Polym. 76, 349–361 (2009) Zielinski, A.A.F., Haminiuk, C.W.I., Nunes, C.A., Schnitzler, E., van Huth, S.M., Granato, D.: Chemical composition, sensory properties, provenance, and bioactivity of fruit juices as assessed by chemometrics: a critical review and guideline. Compr. Rev. Food Sci. Food Saf. 13, 300–316 (2014)
123