Open Access
Add to BibSonomy
Abstract
An efficient method has been developed to identify meat species by using laser-induced breakdown spectroscopy (LIBS). To improve the accuracy and stability of meat species identification, multiplicative scatter correction (MSC) was adopted to first pretreat the spectrum for correction of spectrum scatter. Then the corrected spectra were identified by using the K-nearest neighbor (KNN) model. The results showed that the identification rate improved from 94.17% to 100% and the prediction coefficient of variance (CV) decreased from 5.16% to 0.56%. This means that the accuracy and stability of meat species identification using MSC and LIBS simultaneously improved. In light of the findings, the proposed method can be a valuable tool for meat species identification using LIBS.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Meat products play a significant role in the human diet because they are the major source of protein and are rich in microelements. China is the largest consumer country in the world, and the pork consumption in China was over 54.9 million tons in 2017 [1]. In recent years, interest-driven, unscrupulous manufacturers have been adding cheaper meat species into expensive ones, such as beef adulterated with pork [2, 3]. This adulteration leads to financial, ethical, and health problems. So meat species identification has become one of the research hotspots in recent years. The most commonly used analytical tools for meat detection are polymerase chain reaction [4], liquid chromatography [5], and capillary gel electrophoresis [6]. However, the disadvantages of these methods, such as complex operation, time required, and need for professionals [7, 8] restrict their application. Therefore, it is vital to find a simpler, quicker technology for meat species identification.
Laser-induced breakdown spectroscopy (LIBS), a well-established and powerful optical emission spectroscopy (OES) analytical technique, has a wide range of applications. Because of its advantages, including rapid detection, no pretreatment of samples and real-time analysis [15–19], it has been applied in tissue differentiation [9], environmental monitoring [10, 11], and plant sample detection [12–14]. Up to now, only a few research studies about the recognition of meat species using LIBS have been reported. G. Bilge et al. identified three kinds of meat species using LIBS combined with principal component analysis (PCA), and the accuracy of classification was 83.37% [20]. Fang-Yu Yueh et al. classified six kinds of chicken organs using LIBS combined with hierarchical cluster analysis (HCA), and the identification rate of the tissues was 93.18% [21]. Yi-Ning Zhu et al. combined a support vector machine (SVM) and PCA to improve the discrimination rate and efficiency in LIBS for six kinds of meat and obtained an identification accuracy of 89.11% [22]. However, the studies described above were mainly focused on element information extraction and classifier optimization; they did not involve intrinsic spectral lines correlation, which hampered the accuracy of identification as well as the stability in their work.
In this study, our main aim was to eliminate the spectra scatter effect to improve accuracy and stability of meat identification rather than extract element information directly. We propose multiplicative scatter correction (MSC) as a spectral pretreatment method for scatter correction. The corrected spectra were identified based on the K-nearest neighbor (KNN) model. The test set accuracy and the coefficient of variance were used to evaluate the accuracy and stability of meat identification.
2. Experiments and methods
2.1 Experimental setup and materials
A schematic diagram of the LIBS setup is shown in Fig. 1. In an air environment, the laser beam was obtained by a Q-switched Nd: YAG laser (wavelength: 532 nm; repetition rate: 10 Hz; pulse width: 8 ns; pulse energy: 30 mJ), and it was reflected and then focused onto the surface of the pellet sample by a focal lens (f = 150mm). The plasma emission was collected by a light collector, the signal enters the spectrometer through the optical fiber; the emission spectra were obtained using an echelle spectrometer (spectral range: 200-950 nm; resolution: λ/Δλ = 5000; Andor Tech., Mechelle 5000) coupled with an intensified charge-coupled device (ICCD) camera (Andor Tech., iStar DH-334T). The ICCD was operated in the gated mode, and the gate delay and gate width were set to be 1 and 10 μs to obtain high spectral intensity and signal to noise (SBR).
As shown in Fig. 2. Five kinds of pure meat species (GBW10050 (GSB-28); shrimp, GBW10018(GSB-9); chicken, GBW(E)100197; beef, GBW10024(GSB-15); Scallop, and GBW10051(GSB-29); pig liver) and one kind of mixed sample(shrimp powder into scallop powder with a 1:1 ratio) were used in this work. The meat powder samples analyzed were pressed into pellets form with diameter of 40mm using pressure of 20MPa.
Fig. 2 Pellets of six Samples: (a) scallop, (b) shrimp, (c) pig liver, (d) chicken, (e) beef, and (f) mixed sample
2.2 Algorithm description
MSC is currently a widely used chemometrics method for multi-wavelength calibration modeling, which can effectively eliminate the baseline shift and offset. As a pretreatment for improving the spectrum quality method, MSC has been applied in infrared spectrum for scattering correction [27–30]. M.R. Maleki et al. applied MSC as an important pre-processing step for spectra in an automatic near infrared (NIR) sensor for on-line measurement of some material properties [31], and the scattering effects can be eliminated effectively with MSC. To solve the spectral scatter problem, MSC was proposed in this study. The program was written in MATLAB R2014a. The main steps involved in the MSC method are listed below [29].
3. Results and discussion
3.1 Analytical lines selection
With the experimental setup and spectral acquisition parameters, the LIBS spectrum from 200 to 950nm of six pellets are shown in Fig. 3. The major metallic elements (magnesium (Mg), sodium (Na), potassium (K), calcium (Ca), and aluminum (Al)), and nonmetallic elements (carbon (C), hydrogen (H), oxygen (O), nitrogen (N), and molecular bonds (C-N)) were observed. As a result, the difference in the concentration of the elements in six pellets can be reflected in the spectral intensity of their LIBS spectra, which is the basis for discriminating the meat species effectively.
Fig. 3 LIBS spectrum of meat samples ((a) shrimp, (b) chicken, (c) beef, (d) scallop, (e) pig liver, and (f) mixed sample)
For the purpose of extracting information on the element content and reducing the amount of calculation, spectral lines should be selected. Each spectral line is a coordinate value of a dimension in the KNN model, so it is important to select appropriate characteristic spectral lines for meat species classification. Based on the meat elements contained in the sample, five kinds of metallic elements (magnesium (Mg), sodium (Na), potassium (K), calcium (Ca), and aluminum (Al)), and 5 kinds of nonmetallic elements (carbon (C), hydrogen (H), oxygen (O), nitrogen (N), and molecular bonds (C-N)) were selected as characteristic elements. The spectral lines listed in Table 1 were selected to identify the meat species. In the detectable spectral range of the spectrometer, 31 spectral lines were selected as the KNN input dimension. The principle of selecting characteristic spectral lines should be met as follows:
Table 1. Analytical lines used as input variables for KNN
- (1) The element compositions of Ca, K, Na, Al, and Mg vary depending on meats species [23, 24];
- (2) Those emission line wavelength errors must be within ± 0.1nm relative to the National Institute of Standards and Technology (NIST) database [25].
- (3) The self-absorption or self-reversed characteristic spectral lines will not be selected.
3.2 Meat identification without MSC
For each meat species pellets, 40 spectra were recorded. In total, 240 spectra for six kinds of meat were obtained. The former 20 spectra of each meat species pellets, 120 spectra in total, were used to train the KNN model. The other 120 spectra were used to test the performance of the KNN model and to verify the identification accuracy. The identification results of six kinds of meat species by the KNN model are shown in Fig. 4. The X-axis represents the 120 spectra test set, which was arranged according to the specie of meat. The Y-axis represents the spectral labels that mark meat species. In the axis, (1, shrimp) represents the first spectra as shrimp. The “〇” represents the actual label, while the “*” represents the predicted label by the KNN model. The identification accuracy of the KNN model is 94.17%, the identification accuracy for each meat specie: shrimp, chicken, beef, scallop, pig liver, and the mixed sample were 90%, 95%, 100%, 95%, 85% and 100%, respectively.
The reasons for misidentification of the model above are as follows, when the focused laser ablated the pressed powder sample, powder splash resulted in the spectral scatter that causes noise to produce spectral fluctuations. Since the KNN model is easily affected by noise [26], there will be misidentification between the close- Euclidean distance meat species when the spectral fluctuation is large. The relative position of each spectrum and their average spectrum was calculated by multidimensional scaling (MDS) and is shown in Fig. 5. As the two-dimensional relative position, the closest Euclidean distance of the average spectra of the six kinds of meat species was chicken and pig liver, and the second closest is beef and scallop. The results of the KNN model are in conformity with the distance calculation, and the closer the Euclidean distance of the average spectra is, the higher the possibility of misidentification.
3.3 KNN identification with MSC
With the MSC, the reduction in relative standard deviation (RSD) is shown in Table 2. The average RSD values of the main metal elements Ca, K, Na, Mg, and Al decreased from 26.73%, 11.11%, 12.14%, 29.47%, and 25.07% to 20.13%, 7.50%, 6.39%, 22.91%, and 24.12%, respectively; and the average RSD value of the selected spectral lines decreased from 27.00% to 22.02%. Spectral fluctuation decreased with the reduction in the spectra lines RSD; and for the KNN model, the lower the spectral fluctuations, the higher the identification rate.
Table 2. The average RSD of six kinds of sample at selected spectral lines
The 240 spectra of all six kinds of meat species pellets were processed by MSC. For the train/test form above, the former 120 processed spectra were selected to train the new KNN model. The performance of the new model was tested by the other 120 processed spectra. The identification results of all six kinds of meat pellets by the new KNN model were shown in Fig. 6. After the processing of MSC, the identification rate of the six kinds of meat pellets improved from 94.17% to 100%. The identification rate of each kind of meat species and the average accuracy of classification were shown in Table 3. For the 31 spectral lines, the average distance of pig liver and chicken/beef and scallop was very close. For the KNN model, the closer the two species are, the lower the identification rate. When the spectral fluctuations declined, the average distance of each meat species decreased, and the accuracy of the meat species identification improved. With the decline of spectral fluctuation, pig liver and chicken can be identified completely, as well as beef and scallop.
Table 3. The accuracy of identification with MSC and without MSC
To further study the stability, the data set was divided into a train set and a test set randomly with a 1:1 ratio. The processing was repeated 2,000 times for verification. The average CV decreased from 5.16% to 0.56%.
With the reduction of RSD of elements, the relative position of each spectrum and their average spectrum was calculated by multidimensional scaling (MDS) and is shown in Fig. 7. The spectra relative position gather together to their average spectrum comparing to the relative position without MSC. The beef and scallop can be recognized into two groups with no superposition, and the distance of chicken and pig liver is farther. Therefore, the CV of the prediction model decreased significantly, and the repeatability of meat species recognition improved significantly.
4. Conclusions
The aim of this work was to improve the accuracy and stability of meat species identification by using multiplicative scatter correction combined with K-nearest neighbor in LIBS. With the MSC, the RSD of selected metal elements (Na, K, Mg, and Ca) decreased. The identification accuracy of six kinds of meat species improved from 94.17% to 100%. Furthermore, with the proposed method, the CV decreased from 5.16% to 0.56%. The results showed that the accuracy and stability of meat recognition were improved simultaneously. Therefore, the MSC algorithm is an effective data processing method for LIBS analysis in meat species identification.
Funding
National Natural Science Foundation of China (No. 51429501 and 61575073) and China Postdoctoral Science Foundation(No. 2017M622415).
References and links
1. United States Department of Agriculture, “Livestock and Poultry: World Markets and Trade(2017) ”, https://apps.fas.usda.gov/psdonline/circulars/livestock_poultry.pdf.
2. M. Kamruzzaman, D. W. Sun, G. ElMasry, and P. Allen, “Fast detection and visualization of minced lamb meat adulteration using NIR hyperspectral imaging and multivariate image analysis,” Talanta 103, 130–136 (2013). [CrossRef] [PubMed]
3. X. Tian, J. Wang, and S. Cui, “Analysis of pork adulteration in minced mutton using electronic nose of metal oxide sensors,” J. Food Eng. 119(4), 744–749 (2013). [CrossRef]
4. M. A. Rodríguez, T. García, I. González, L. Asensio, P. E. Hernández, and R. Martín, “Quantitation of mule duck in goose foie gras using TaqMan real-time Polymerase Chain Reaction,” J. Agric. Food Chem. 52(6), 1478–1483 (2004). [CrossRef] [PubMed]
5. P. A. Guy, M. C. Savoy, and R. H. Stadler, “Quantitative analysis of clenbuterol in meat products using liquid chromatography-electrospray ionisation tandem mass spectrometry,” J. Chromatogr. B Biomed. Sci. Appl. 736(1-2), 209–219 (1999). [CrossRef] [PubMed]
6. B. Vallejo-Córdoba and M. Cota-Rivas, “Meat species identification by linear discriminant analysis of capillary electrophoresis protein profiles,” J. Capillary Electrophor. 5(5-6), 171–175 (1998). [PubMed]
7. A. Kumar, F. Y. Yueh, J. P. Singh, and S. Burgess, “Characterization of malignant tissue cells by laser-induced breakdown spectroscopy,” Appl. Opt. 43(28), 5399–5403 (2004). [CrossRef] [PubMed]
8. L. Liu, L. Deng, L. Fan, X. Huang, Y. Lu, X. Shen, L. Jiang, J. F. Silvain, and Y. Lu, “Time-resolved resonance fluorescence spectroscopy for study of chemical reactions in laser-induced plasmas,” Opt. Express 25(22), 27000–27007 (2017). [CrossRef] [PubMed]
9. R. Kanawade, F. Mahari, F. Klämpfl, M. Rohde, C. Knipfer, K. Tangermann-Gerk, W. Adler, M. Schmidt, and F. Stelzle, “Qualitative tissue differentiation by analysing the intensity ratios of atomic emission lines using laser induced breakdown spectroscopy (LIBS): prospects for a feedback mechanism for surgical laser systems,” J. Biophotonics 8(1-2), 153–161 (2015). [CrossRef] [PubMed]
10. D. Girón, T. Delgado, J. Ruiz, L. Cabalín, and J. Laserna, “In-situ monitoring and characterization of airborne solid particles in the hostile environment of a steel industry using stand-off LIBS,” Measurement 115, 1–10 (2018). [CrossRef]
11. W. A. Farooq, W. Tawfik, F. N. Almutairi, and Z. A. Alahmed, “Qualitative Analysis and Plasma Characteristics of Soil from a Desert Area using LIBS Technique,” J. Opt. Soc. Korea 17(6), 548–558 (2013). [CrossRef]
12. K. Devey, M. Mucalo, G. Rajendram, and J. Lane, “Pasture Vegetation Elemental Analysis by Laser Induced Breakdown Spectroscopy,” Commun. Soil Sci. Plant Anal. 46(sup1), 72–80 (2015). [CrossRef]
13. J. N. Kunz, D. V. Voronine, H. W. Lee, A. V. Sokolov, and M. O. Scully, “Rapid detection of drought stress in plants using femtosecond laser-induced breakdown spectroscopy,” Opt. Express 25(7), 7251–7262 (2017). [CrossRef] [PubMed]
14. L. Krajcarová, K. Novotný, M. Galiová, V. Kanický, V. Adam, and R. Kizek, “Mapping of elements distribution in plant samples using LIBS,” Elemental mapping; laser induced breakdown spectroscopy; LIBS; plants (2010).
15. W. B. Lee, J. Wu, Y. I. Lee, and J. Sneddon, “Recent Applications of Laser‐Induced Breakdown Spectrometry: A Review of Material Approaches,” Appl. Spectrosc. Rev. 39(1), 27–97 (2004). [CrossRef]
16. Z. Wang, T.-B. Yuan, Z.-Y. Hou, W.-D. Zhou, J.-D. Lu, H.-B. Ding, and X.-Y. Zeng, “Laser-induced breakdown spectroscopy in China,” Front. Phys. China 9, 419–438 (2014).
17. L. B. Guo, Z. Q. Hao, M. Shen, W. Xiong, X. N. He, Z. Q. Xie, M. Gao, X. Y. Li, X. Y. Zeng, and Y. F. Lu, “Accuracy improvement of quantitative analysis by spatial confinement in laser-induced breakdown spectroscopy,” Opt. Express 21(15), 18188–18195 (2013). [CrossRef] [PubMed]
18. X. Cheng, X. Yang, Z. Zhu, L. Guo, X. Li, Y. Lu, and X. Zeng, “On-stream analysis of iron ore slurry using laser-induced breakdown spectroscopy,” Appl. Opt. 56(33), 9144–9149 (2017). [CrossRef] [PubMed]
19. Y. M. Guo, L. M. Deng, X. Y. Yang, J. M. Li, K. H. Li, Z. H. Zhu, L. B. Guo, X. Y. Li, Y. F. Lu, and X. Y. Zeng, “Wavelet-based interference correction for laser-induced breakdown spectroscopy,” J. Anal. At. Spectrom. 32(12), 2401–2406 (2017). [CrossRef]
20. G. Bilge, H. M. Velioglu, B. Sezer, K. E. Eseller, and I. H. Boyaci, “Identification of meat species by using laser-induced breakdown spectroscopy,” Meat Sci. 119, 118–122 (2016). [CrossRef] [PubMed]
21. F. Y. Yueh, H. Zheng, J. P. Singh, and S. Burgess, “Preliminary evaluation of laser-induced breakdown spectroscopy for tissue classification,” Spectrochim. Acta B At. Spectrosc. 64(10), 1059–1067 (2009). [CrossRef]
22. Y.-N. Zhu, Y. Ping, and X.-Y. Yang, “Classification of Fresh Meat Species Using Laser-induced Breakdown Spectroscopy with Support Vector Machine and Principal Component Analysis,” Chin. J. Anal. Chem. 45, 336–341 (2017).
23. C. E. Bodwell, and B. A. ANDERSON, Nutritional Composition and Value of Meat and Meat Products (1986).
24. G. Lombardi-Boccia, S. Lanzi, and A. Aguzzi, “Aspects of meat quality: trace elements and B vitamins in raw and cooked meats,” J. Food Compos. Anal. 18(1), 39–46 (2005). [CrossRef]
25. NIST, “NIST Atomic Spectra Database”, retrieved Oct. 2014, http://www.nist.gov/pml/data/asd.cfm
26. Y. Jiang and Z. H. Zhou, “Editing training data for kNN classifiers with neural network ensemble,” Advances in Neural Networks–ISNN 2004, 356–361 (2004).
27. Y. Liu, X. Sun, and A. Ouyang, “Nondestructive measurement of soluble solid content of navel orange fruit by visible–NIR spectrometric technique with PLSR and PCA-BPNN,” Lebensm. Wiss. Technol. 43(4), 602–607 (2010). [CrossRef]
28. H. Chen, Q. Song, G. Tang, Q. Feng, and L. Lin, “The combined optimization of Savitzky-Golay smoothing and multiplicative scatter correction for FT-NIR PLS models,” ISRN Spectroscopy 2013 (2013). [CrossRef]
29. T. Isaksson and T. Næs, “The effect of multiplicative scatter correction (MSC) and linearity improvement in NIR spectroscopy,” Appl. Spectrosc. 42(7), 1273–1284 (1988). [CrossRef]
30. I. S. Helland, T. Næs, and T. Isaksson, “Related versions of the multiplicative scatter correction method for preprocessing spectroscopic data,” Chemom. Intell. Lab. Syst. 29(2), 233–241 (1995). [CrossRef]
31. M. R. Maleki, A. M. Mouazen, H. Ramon, and J. D. Baerdemaeker, “Multiplicative Scatter Correction during On-line Measurement with Near Infrared Spectroscopy,” Biosyst. Eng. 96(3), 427–433 (2007). [CrossRef]
