Open Access
Add to BibSonomy
Abstract
Nutrient profile determination for plant materials is an important task to determine the quality and safety of the human diet. Laser-induced breakdown spectroscopy (LIBS) is an atomic emission spectrometry of the material component analytical technique. However, quantitative analysis of plant materials using LIBS usually suffers from matrix effects and nonlinear self-absorption. To overcome this problem, a hybrid quantitative analysis model of the partial least squares-artificial neural network (PLS-ANN) was used to detect the compositions of plant materials in the air. Specifically, fifty-eight plant materials were prepared to split into calibration, validation and prediction sets. Nine nutrient composition profiles of Mg, Fe, N, Al, B, Ca, K, Mn, and P were employed as the target elements for quantitative analysis. It demonstrated that the prediction ability can be significantly improved by the use of the PLS-ANN hybrid model compared to the method of standard calibration. Take Mg and K as examples, the root-mean-square errors of calibration (RMSEC) of Mg and K were decreased from 0.0295 to 0.0028 wt.% and 0.2884 to 0.0539 wt.%, and the mean percent prediction errors (MPE) were decreased from 5.82 to 4.22% and 8.82 to 4.12%, respectively. This research provides a new way to improve the accuracy of LIBS for quantitative analysis of plant materials.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Corrections
Ping Yang, Xiangyou Li, and Zhanglong Nie, "Determination of nutrient profile in plant materials using laser-induced breakdown spectroscopy with partial least squares-artificial neural network hybrid model: erratum," Opt. Express 29, 20687-20687 (2021)https://opg.optica.org/oe/abstract.cfm?uri=oe-29-13-20687
1. Introduction
Macronutrients (N, P, K, Ca, Mg, etc.) and micronutrients (Fe, Mn, Zn, B, etc) play a decisive role in plant nutrition and can affect crop yields when not present in appropriate concentration levels. Nutrient profile deficiency can cause severe problems in plant growth and appearance. If high yields and ideally effective production levels are to be maintained, it is necessary to determine the essential elements in plant materials to evaluate the nutritional status of economic crops.
Conventional analytical techniques, such as atomic absorption spectrometry (AAS) [1], inductively coupled plasma mass spectrometry (ICP-MS) [2], and inductively coupled plasma optical emission spectrometry (ICP-OES) [3], have been applied in essential elements determination in agricultural product. These methods have obtained good detection limits, sensitivity, and stability. However, almost all these techniques produce large amounts of toxic waste, and require the consumption of chemical reagents for digesting samples, which are time-consuming, complex to operate, and may introduce pollutants into ecosystems. Therefore, it is still a challenging work to seek a rapid, eco-friendly and accurate detection method for agricultural product.
Laser-induced breakdown spectroscopy (LIBS) is a spectroscopic technique for material composition analysis [4–6]. With the characteristics of minimal sample preparation, rapid analyzing speed and multi-elemental analysis, it is a useful tool for rapid, real-time and in-situ measurements for qualitative or quantitative analysis, and has been widely applied in industrial production [7–10], environmental monitoring [11–14], food safety monitoring [15–19], and other fields [20–23]. Some researchers have carried out extensive works on agricultural detection. For example, De Carvalho et al. [24] employed fs-LIBS with partial least squared (PLS) in the determination of nutrient profile in plant materials. The prediction errors of below 20% on future unknown samples are expected. Yao et al. [25] used LIBS coupled with PLS to detect Cd in polluted fresh leafy vegetables, and obtained the correlation coefficient of 0.9800 and root-mean-square errors of prediction (RMSEP) of 0.6714 mg/kg in the prediction set. Peng et al. [26] used LIBS coupled with PLS to detect Cr in rice leaves, and achieved the correlation coefficient of 0.9669 and RMSEP of 4.75 mg/kg in the prediction set. Sirven et al. [27] applied PLS and artificial neural network (ANN) model to analyze Cr in soil samples using LIBS, respectively. They found that the prediction accuracy of ANN model is higher than PLS model, and the average relative standard deviation are around 10% by ANN model. Mukhono et al. [28] adopted PLS and ANN model to analyze As, Cr, Cu, Pb, and Ti in soil and rock samples using LIBS. Their results show that PLS model is superior to the ANN model in soil quantitative analysis, but ANN model is better than the PLS model in rock analysis. The ANN model is more robust than the PLS model at modeling spectral nonlinearities and correcting matrix effects. Above studies also indicate that the prediction accuracy of the PLS and ANN models depends on specific samples. Therefore, combining the advantages of reducing the multiple collinearity of the independent variable of the PLS and the nonlinear processing capability of the ANN is expected to improve the accuracy of LIBS quantitative analysis. At present, there are few reports on the quantitative analysis for nutrient profile of plant materials using LIBS with PLS-ANN hybrid model.
The determination stability and sensitivity of agricultural product using LIBS are still unsatisfactory due to matrix effects and nonlinear self-absorption. To improve the quantitative accuracy of plant material analysis using LIBS, a PLS-ANN hybrid model was proposed in this work. To deal with the multiple collinearity of independent variable, PLS was used to select latent spectral variables and reduce data dimensions. To deal with matrix effects and nonlinear self-absorption, the ANN model was built using the PLS generated latent variables to quantify essential elements. Calibration models for major components, Mg, Fe, N, Al, B, Ca, K, Mn, and P in plant material samples were established based on proposed PLS-ANN hybrid model.
2. Model description
PLS [4,29,30] is a multivariable linear regression method. It is used to find the relations between the intensity of emission spectral line and the analyte concentrations variables in LIBS. The number of variables in the LIBS spectrum is usually hundreds or thousands, and contains a lot of irrelevant and redundant variables. The full spectrum was usually taken as the input of the PLS model. The irrelevant and redundant variables will influence the prediction ability and stability of the regression model. ANN [27,28,31] is a nonlinear multivariable regression methods. Compared with PLS methods, ANN model can establish a nonlinear mapping relationship between independent variables and dependent variables. It can deal with nonlinear effects caused by self-absorption and correct matrix effects to some extent. In this work, PLS is used to select latent spectral variables and reduce data dimensions, while the ANN model uses the PLS generated latent variables to quantify plant materials.
The flow diagram is described in Fig. 1. The steps involved in the proposed PLS-ANN approach are outlined below:
Step (1): The plant material samples are divided into calibration, validation and prediction sets. The ratio of sample number is approximately 2:1:1. Normalize the spectral intensity matrices of the calibration, validation, and prediction set samples to obtain XC, XV, and XP, respectively. At the same time, the analytical element concentration matrices of the calibration and verification set samples to obtain CC and CV, respectively.
Step (2): Establishing a PLS model based on the normalized spectral intensity matrix XC and the analytical element concentration matrix CC by calibration set samples. The principal component h is determined by the validation set (XV and CV).
Step (3): The PLS model is re-established according to the explanatory variable XC, the reaction variable CC and the principal component n (n = [h-i:h + i]; i=5). Then the weight matrix WC and the principal component matrix TC of the normalized spectral intensity matrix XC are obtained, which satisfies the relation T = XWC.
Step (4): In the case of the unchanged hidden layer and output layer, use the Sigmoid function [29] to train the ANN network according to TC and CC under different principal component n, the learning rate is set to 0.01. The principal components corresponding to the minimum predicted residual error sum of squares (PRESS) is the best principal component m, which is chosen as the ANN's inputs. In order to prevent the under-fitting of the ANN model, the root-mean-square errors of calibration (RMSEC) value of the training set sample is set to 0, and the maximum number of iterations is set to 1000. In order to prevent the over-fitting of the ANN model, the network model of each iteration training is used to predict the concentration of the verification set sample, and the corresponding predicted concentration’s root mean square errors of validation (RMSEV) are obtained. When the RMSEV has a minimum RMSEVmin and then 10 consecutive trainings network model’s RMSEV larger than RMSEVmin, the iteration is terminated as well. The RMSEVmin and its corresponding weights and offsets are recorded.
Step (5): In order to avoid the impact of the initial weight and offset on the ANN model, repeat step (4) 1000 times, and then find the minimum value of 1000 RMSEVmin. The minimum value corresponds to the best ANN model, and the weights and offsets corresponding to the best model are the best values. Under the best principal component m, determine the best ANN network parameter NetC based on TV and CV.
Step (6): Model prediction: the normalized spectral intensity matrix XP corresponding to principal component matrix TP is used as the input of the ANN model. According to the best ANN network model established in the above steps, the predicted concentration matrix C'P can be obtained. By inversely normalizing C'P, the normalized spectral intensity matrix XP corresponding to the predicted concentration CP of can be obtained.
In this work, PLS and ANN combined model was based on the toolbox in MATLAB (version R2015; MathWorks Corp., Natick, Massachusetts, USA).
3. Experimental setup and sample preparation
3.1 Experimental setup
The schematic diagram of the experimental setup used in this work is shown in Fig. 2(a). A Q-switched Nd:YAG pulsed laser (Quantel Brilliant B, maximum energy: 400 mJ/pulse, wavelength: 532 nm, pulse width: 5 ns, repetition rates: 10 Hz) was used to ablate samples. The spot size of focused laser pulse ranges from 500 μm. The laser beam was reflected by a mirror, and focused at 1.25 mm below the sample surface by a UV-grade quartz lens with the focal length of 100 mm. The plasma emission was coupled into an echelle spectrometer (Andor Tech., Mechelle 5000, spectral range from 200 to 950 nm with a resolution of λ/Δλ = 5000, wavelength accuracy was ± 0.05 nm) attached with an intensified charge coupled device (ICCD) camera (Andor Tech., iStar DH-334T). The acquisition and analysis of data were performed using a personal computer. To provide a fresh surface for each laser ablation, the target was mounted on a X-Y-Z motorized translation stage and moving at a speed of 4 mm/s during ablation.
The LIBS spectra were acquired under the following optimal conditions. The laser pulse energy was 40 mJ. The laser frequency was 10 Hz. The gate delay and gate width were 1.5 μs and 3 μs, respectively. To reduce the intensity deviation, each spectrum was accumulated 60 shots. With the help of X-Y-Z motorized translation stage, 10 spectra were recorded for each plant material sample.
3.2 Sample pretreatment process
The following certified reference materials were used in this study: bean (State Administration of Quality Supervision, Inspection, and Quarantine of China - GBW10021), laver (GBW10023), celery (GBW10048), welsh onion (GBW10049). For convenience, the samples were numbered from 1 to 4. The sample information is listed in Table 1. Because small sample numbers result in overtraining and less robust of calibration model, four samples are not sufficient to divide into a model set and a test set for the PLS, ANN and PLS-ANN method. Therefore, a set of mixed samples were created by mixing every two samples (e.g., No.1 and 2, 1 and 3, 1 and 4, 2 and 3, 2 and 4, 3 and 4) with the same matrix in Table 1 by nine ratios (9:1, 8:2, 7:3, 6:4, 5:5, 4:6, 3:7, 2:8, and 1:9). Finally, 54 mixed samples were obtained and were numbered from 5 to 58 respectively. The concentrations of mixed samples can be calculated according to the ratios. The samples were passed through an 80 mesh sieve. Then mixed it until they were homogeneous. To obtain a uniform surface for laser ablation, all the samples were pressed into pellets (Fig. 2(b)) with a pressure of 25 MPa. The schematic diagram of pellet method was shown in our former works [19,32].
Table 1. The major compositions and concentrations of plant material samples
3.3 Model evaluation
To evaluate the prediction of the PLS-ANN hybrid model, the values of determination coefficient (R2), RMSEC, RMSEV, RMSEP, and the mean prediction error (MPE) were used to evaluate the analytical performance [33–35]. The RMSEC, RMSEV, RMSEP, and MPE were described as follows:
4. Results and discussion
4.1 Selection of characteristic spectral lines
The LIBS spectra of plant material samples are very complex. The obtained LIBS spectra contained more than 25,025 wavelength channels in a wavelength range starting at 200 nm in the ultraviolet (UV) and extending into the near-infrared (NIR) to 950 nm.
Figure 3 shows the plasma emission spectra of bean. The characteristic lines of the analytical elements in the sample are obtained by comparing the standard atomic spectrum database of the National Institute of Standards and Technology (NIST). In detail, most of the valuable and high-intensity spectral lines were located around the regions of 247∼255 nm, 370∼380 nm, 390∼405 nm, 515∼520 nm, and 740∼775 nm, respectively. Figure 3 showed the LIBS spectrum of bean which clearly showed the presence of prominent lines of Mg, Fe, N, Al, B, Ca, K, Mn, and P. The emission lines standing for different elements were accurately identified and tagged in corresponding positions. The spectral lines of analytical elements are listed in Table 2.
Table 2. The compositions of plant material, corresponding characteristic wavelengths and spectral regions
4.2 Quantitative analysis using the standard calibration
Traditional quantitative analysis method was mainly based on the relationship between the intensity of emission spectral line and concentrations of analyte. Figure 3 showed the typical LIBS spectra of plant material. The spectral lines of Mg, Fe, N, Al, B, Ca, K, Mn, and P were detected in the analyzed plant material samples. In this study, the emission lines Mg I 518.36 nm, Fe I 373.71 nm, N I 746.83 nm, Al I 396.15nm, B I 249.77 nm, Ca I 396.85 nm, K I 769.89 nm, Mn I 403.08 nm, and P I 253.56 nm were chosen to establish the quantitative models. In order to reduce the fluctuation of spectral data, each spectrum was normalized by the Si I 288.16 nm at the corresponding region.
Take Mg and K as examples, Fig. 4 showed the intensities of Mg I 518.36 nm and K I 769.89 nm varying with the concentrations in plant materials. The R2, RMSEC, and MPE value of Mg were 0.8577, 0.0295 wt.%, and 5.82%, respectively. The R2, RMSEC, and MPE values of K were 0.5451, 0.2884 wt.%, 8.82%, respectively. It clearly showed that there was no satisfactory collinearity between the intensity and the concentrations of the nutrient profile for all the samples using the standard calibration model due to the physical and chemical matrix effects. The R2, RMSEC, and MPE of other nutrient profile are listed in Table 3. The analysis results cannot meet the requirement of LIBS applications. Therefore, more effective approaches should be developed for the plant material nutrient profile determination.
Fig. 4. Standard calibration curves of Mg I 518.36 nm/ Si I 288.16 nm (a) and K I 769.89 nm / Si I 288.16 nm (b).
Table 3. Quantitative analysis results with standard calibration
4.3 Quantitative analysis using the PLS-ANN hybrid model
To further improve the prediction accuracy of the essential elements content, PLS-ANN hybrid models were introduced for the determination of nutrient profile. In this adaptive model, PLS is used to select latent spectral variables and reduce data dimensions, while a ANN model uses the PLS generated latent variables to quantify essential elements. Combination of PLS and ANN is a promising modeling approach to ensuring representative spectral variables being selected, and the matrix interference and nonlinear self-absorption effect can be effectively managed using ANN.
To analyze the concentrations of Mg, Fe, N, Al, B, Ca, K, Mn, and P, nine typical wavelength regions (see in Table 2) containing the emission lines of corresponding elements were chosen for each sample. Among the 58 plant material samples, 29 samples (No. 1, 3, … 57) were used as a model set to construct the calibration model. The other 15 samples (No. 2, 6, … 58) were used as a test set to validate the model. The remaining 14 samples (No. 4, 8, … 56) were used as a test set to predict the model. Before the spectral data were imported to the program, each spectrum was normalized by the Si I 288.16 nm at the corresponding region.
So as to evaluate the prediction ability of the PLS, ANN, and PLS-ANN hybrid model, the analytical performance of nine essential elemental components in plant materials samples were compared. Taking Mg as an example, the quantitative analysis results are shown in Fig. 5. Standard accuracy is evaluated by the RMSEC (including calibration and validation sets). Prediction accuracy is evaluated by RMSEP and MPE. The effectiveness of the model should be evaluated by the predictive ability of the model. The RMSEC of PLS, ANN, and PLS-ANN model were 0.0179, 0.0028, and 0.0028 wt.%, respectively. The RMSEP and MPE were 0.0321, 0.0203, 0.0231 wt.% and 6.15, 4.27, 4.22%, respectively. The comparison results show that the PLS-ANN hybrid model has the highest accuracy, the ANN model is the second, and the PLS model is the lowest. PLS is a multivariable linear regression method. It means that each spectral variable corresponds to a constant coefficient in the PLS model. Therefore, the performance of essential elemental analysis using LIBS with PLS is unsatisfactory due to nonlinear self-absorption and matrix effects. The prediction accuracy of the PLS-ANN hybrid model is higher than the PLS model, indicating that the relationship between the principal component of the spectral intensity in the plant material and the concentration of the analytical element tends to be nonlinear. When the principal component is extracted by the PLS method, the ANN method can better simulate the nonlinear mapping relationship. The principal component is used as the independent variable of the ANN model. This independent variable overcomes the multiple collinearity of the original variable and filters the noise of the original data, which can establish the ANN model more effectively. Therefore, the prediction accuracy of the PLS-ANN hybrid model is better than the ANN model.
Fig. 5. Calibration regression curve of the Mg by the PLS model (a), ANN model (b), and PLS-ANN model (c).
The analytical performance of the remaining eight components (Fe, N, Al, B, Ca, K, Mn, and P) by PLS, ANN, and PLS-ANN hybrid model is also summarized in Table 4. The shading in the Table 4 shows the best data among the three models. The RMSEC values of the PLS-ANN hybrid models of all the analytical components are the lowest, indicating that the PLS-ANN hybrid model has the best calibration results. The RMSEP and MPE values of the PLS-ANN hybrid model have eight minimum value among the nine elemental components. The results show that the PLS-ANN hybrid model can significantly improve the prediction accuracy.
Table 4. Comparison of quantitative analysis results with different models.
5. Conclusions
In this work, to overcome the matrix effects and nonlinear self-absorption on quantitative analysis, PLS-ANN hybrid model was proposed to improve the analysis accuracy of plant materials using LIBS. PLS was used for variable selection, and ANN was used to establish the nonlinearity regression model between spectral data and concentrations in the PLS-ANN hybrid model. Combined with the advantages of reducing the multiple collinearity of PLS independent variables and the nonlinear processing ability of ANN, the accuracy of LIBS quantitative analysis is significantly improved. With the hybrid model, the RMSEP values of Mg, Fe, N, Al, B, Ca, K, Mn, and P were 0.0231, 0.0117, 0.1011, 0.0409, 0.0003, 0.1541, 0.1431, 0.0006, and 0.0205 wt.%, respectively; the MPE values were 4.22, 13.29, 2.61, 19.02, 8.72, 12.37, 4.12, 7.91, and 3.87%, respectively. The results suggest that the hybrid model is a competitive data processing method for plant material analysis using LIBS.
Funding
University-level scientific research projects of Changzhou College of Information Technology (CXKZ201909Z, CXKZ201910Q); National Natural Science Foundation of China (11874167); High-level Key Professional Construction Projects of Jiangsu vocational colleges (SJG201717131); Qinglan Project of Jiangsu Province of China (GB20190342).
Disclosures
The authors declare no conflicts of interest.
References
1. B. B. Basak and R. Nagaraja Reddy, “Elemental Analysis of Different Cultivar of Senna (Cassia angustifolia) using Microwave Digestion and Flame Atomic Absorption Spectroscopy,” Natl. Acad. Sci. Lett. 40(4), 245–248 (2017). [CrossRef]
2. R. D’Amato, M. Petrelli, P. Proietti, A. Onofri, L. Regni, D. Perugini, and D. Businelli, “Determination of changes in the concentration and distribution of elements within olive drupes (cv. Leccino) from Se biofortified plants, using laser ablation inductively coupled plasma mass spectrometry,” J. Sci. Food Agric. 98(13), 4971–4977 (2018). [CrossRef]
3. J. Chirinos, D. Oropeza, J. Gonzalez, V. Zorba, and R. E. Russo, “Analysis of Plant Leaves Using Laser Ablation Inductively Coupled Plasma Optical Emission Spectrometry: Use of Carbon to Compensate for Matrix Effects,” Appl. Spectrosc. 71(4), 709–720 (2017). [CrossRef]
4. P. Pořízka, J. Klus, E. Képeš, D. Prochazka, D. W. Hahn, and J. Kaiser, “On the utilization of principal component analysis in laser-induced breakdown spectroscopy data analysis, a review,” Spectrochim. Acta, Part B 148, 65–82 (2018). [CrossRef]
5. D. W. Hahn and N. Omenetto, “Laser-induced breakdown spectroscopy (LIBS), part I: review of basic diagnostics and plasma–particle interactions: still-challenging issues within the analytical plasma community,” Appl. Spectrosc. 64(12), 335A–336A (2010). [CrossRef]
6. D. W. Hahn and N. Omenetto, “Laser-Induced Breakdown Spectroscopy (LIBS), Part II: Review of Instrumental and Methodological Approaches to Material Analysis and Applications to Different Fields,” Appl. Spectrosc. 66(4), 347–419 (2012). [CrossRef]
7. L. X. Sun, H. B. Yu, Z. B. Cong, H. Lu, B. Cao, P. Zeng, W. Dong, and Y. Li, “Applications of laser-induced breakdown spectroscopy in the aluminum electrolysis industry,” Spectrochim. Acta, Part B 142, 29–36 (2018). [CrossRef]
8. P. Pořízka, J. Klus, D. Prochazka, E. Képeš, A. Hrdlička, J. Novotný, K. Novotný, and J. Kaiser, “Laser-Induced Breakdown Spectroscopy coupled with chemometrics for the analysis of steel: The issue of spectral outliers filtering,” Spectrochim. Acta, Part B 123, 114–120 (2016). [CrossRef]
9. S. C. Yao, J. L. Xu, J. B. Zhao, K. J. Bai, J. D. Lu, and Z. M. Lu, “Characterization of Fly Ash Laser-Induced Plasma for Improving the On-line Measurement of Unburned Carbon in Gas-Solid Flow,” Energy Fuels 31(5), 4681–4686 (2017). [CrossRef]
10. Z. Q. Hao, L. Liu, R. Zhou, Y. W. Ma, X. Y. Li, L. B. Guo, Y. F. Lu, and X. Y. Zeng, “One-point and multi-line calibration method in laser-induced breakdown spectroscopy,” Opt. Express 26(18), 22926–22933 (2018). [CrossRef]
11. R. X. Yi, X. Y. Yang, R. Zhou, J. M. Li, H. W. Yu, Z. Q. Hao, L. B. Guo, X. Y. Li, Y. F. Lu, and X. Y. Zeng, “Determination of Trace Available Heavy Metals in Soil Using Laser-Induced Breakdown Spectroscopy Assisted with Phase Transformation Method,” Anal. Chem. 90(11), 7080–7085 (2018). [CrossRef]
12. O. Krystofova, V. Shestivska, M. Galiova, K. Novotny, J. Kaiser, J. Zehnalek, P. Babula, R. Opatrilova, V. Adam, and R. Kizek, “Sunflower Plants as Bioindicators of Environmental Pollution with Lead (II) Ions,” Sensors 9(7), 5040–5058 (2009). [CrossRef]
13. M. Y. Yao, J. L. Lin, M. H. Liu, and Y. Xu, “Detection of chromium in wastewater from refuse incineration power plant near Poyang Lake by laser induced breakdown spectroscopy,” Appl. Opt. 51(10), 1552–1557 (2012). [CrossRef]
14. R. X. Yi, L. B. Guo, X. H. Zou, J. M. Li, Z. Q. Hao, X. Y. Yang, X. Y. Li, X. Y. Zeng, and Y. F. Lu, “Background removal in soil analysis using laser- induced breakdown spectroscopy combined with standard addition method,” Opt. Express 24(3), 2607–2618 (2016). [CrossRef]
15. G. Kim, J. Kwak, J. Choi, and K. Park, “Detection of nutrient elements and contamination by pesticides in spinach and rice samples using laser-induced breakdown spectroscopy (LIBS),” J. Agric. Food Chem. 60(3), 718–724 (2012). [CrossRef]
16. H. Jull, R. Künnemeyer, and P. Schaare, “Nutrient quantification in fresh and dried mixtures of ryegrass and clover leaves using laser-induced breakdown spectroscopy,” Precision Agriculture (2018).
17. B. Sezer, H. Apaydin, G. Bilge, and I. H. Boyaci, “Coffee arabica adulteration: Detection of wheat, corn and chickpea,” Food Chem. 264, 142–148 (2018). [CrossRef]
18. P. Yang, R. Zhou, W. Zhang, R. X. Yi, S. S. Tang, L. B. Guo, Z. Q. Hao, X. Y. Li, Y. F. Lu, and X. Y. Zeng, “High-sensitivity determination of cadmium and lead in rice using laser-induced breakdown spectroscopy,” Food Chem. 272, 323–328 (2019). [CrossRef]
19. P. Yang, Y. N. Zhu, X. Y. Yang, J. M. Li, S. S. Tang, Z. Q. Hao, L. B. Guo, X. Y. Li, X. Y. Zeng, and Y. F. Lu, “Evaluation of sample preparation methods for rice geographic origin classification using laser-induced breakdown spectroscopy,” J. Cereal Sci. 80, 111–118 (2018). [CrossRef]
20. A. Hrdlička, J. Hegrová, K. Novotný, V. Kanický, D. Prochazka, J. Novotný, P. Modlitbová, L. Sládková, P. Pořízka, and J. Kaiser, “Sulfur determination in concrete samples using laser-induced breakdown spectroscopy and limestone standards,” Spectrochim. Acta, Part B 142, 8–13 (2018). [CrossRef]
21. M. A. Gondal, M. A. Shemis, A. A. Khalil, M. M. Nasr, and B. Gondal, “Laser produced plasma diagnosis of carcinogenic heavy metals in gallstones,” J. Anal. At. Spectrom. 31(2), 506–514 (2016). [CrossRef]
22. J. Wang, L. Li, P. Yang, Y. Chen, Y. N. Zhu, M. Tong, Z. Q. Hao, and X. Y. Li, “Identification of cervical cancer using laser-induced breakdown spectroscopy coupled with principal component analysis and support vector machine,” Lasers Med. Sci. 33(6), 1381–1386 (2018). [CrossRef]
23. J. Novotný, K. Novotný, D. Prochazka, A. Hrdlicka, and J. Kaiser, “Two dimensional elemental mapping by laser-induced breakdown spectroscopy,” Spectrosc. Eur 26, 6–10 (2014).
24. G. G. A. de Carvalho, J. Moros, D. Santos, F. J. Krug, and J. J. Laserna, “Direct determination of the nutrient profile in plant materials by femtosecond laser-induced breakdown spectroscopy,” Anal. Chim. Acta 876, 26–38 (2015). [CrossRef]
25. M. Yao, H. Yang, L. Huang, T. Chen, G. Rao, and M. Liu, “Detection of heavy metal Cd in polluted fresh leafy vegetables by laser-induced breakdown spectroscopy,” Appl. Opt. 56(14), 4070–4075 (2017). [CrossRef]
26. J. Peng, Y. He, L. Ye, T. Shen, F. Liu, W. Kong, X. Liu, and Y. Zhao, “Moisture Influence Reducing Method for Heavy Metals Detection in Plant Materials Using Laser-Induced Breakdown Spectroscopy: A Case Study for Chromium Content Detection in Rice Leaves,” Anal. Chem. 89(14), 7593–7600 (2017). [CrossRef]
27. J. B. Sirven, B. Bousquet, L. Canioni, and L. Sarger, “Laser-induced breakdown spectroscopy of composite samples: comparison of advanced chemometrics methods,” Anal. Chem. 78(5), 1462–1469 (2006). [CrossRef]
28. P. M. Mukhono, K. H. Angeyo, A. Dehayem-Kamadjeu, and K. A. Kaduki, “Laser induced breakdown spectroscopy and characterization of environmental matrices utilizing multivariate chemometrics,” Spectrochim. Acta, Part B 87, 81–85 (2013). [CrossRef]
29. K. S. Song, L. Li, S. Li, L. Tedesco, H. T. Duan, Z. C. Li, K. Shi, J. Du, Y. Zhao, and T. T. Shao, “Using Partial Least Squares-Artificial Neural Network for Inversion of Inland Water Chlorophyll-a,” IEEE Trans. Geosci. Electron. 52(2), 1502–1517 (2014). [CrossRef]
30. P. Yu, M. Y. Low, and W. Zhou, “Development of a partial least squares-artificial neural network (PLS-ANN) hybrid model for the prediction of consumer liking scores of ready-to-drink green tea beverages,” Food Res. Int. 103, 68–75 (2018). [CrossRef]
31. H. W. Kuhn and A. W. Tucker, “Nonlinear programming,” InTraces and emergence of nonlinear programming, 247–258 (2014).
32. P. Yang, Y. N. Zhu, S. S. Tang, Z. Q. Hao, L. B. Guo, X. Y. Li, Y. F. Lu, and X. Y. Zeng, “Analytical-performance improvement of laser-induced breakdown spectroscopy for the processing degree of wheat flour using a continuous wavelet transform,” Appl. Opt. 57(14), 3730 (2018). [CrossRef]
33. X. H. Zou, L. B. Guo, M. Shen, X. Y. Li, Z. Q. Hao, Q. D. Zeng, Y. F. Lu, Z. M. Wang, and X. Y. Zeng, “Accuracy improvement of quantitative analysis in laser-induced breakdown spectroscopy using modified wavelet transform,” Opt. Express 22(9), 10233–10238 (2014). [CrossRef]
34. K. H. Li, L. B. Guo, J. M. Li, X. Y. Yang, R. X. Yi, X. Y. Li, Y. F. Lu, and X. Y. Zeng, “Quantitative analysis of steel samples using laser-induced breakdown spectroscopy with an artificial neural network incorporating a genetic algorithm,” Appl. Opt. 56(4), 935–941 (2017). [CrossRef]
35. J. El Haddad, L. Canioni, and B. Bousquet, “Good practices in LIBS analysis: Review and advices,” Spectrochim. Acta, Part B 101, 171–182 (2014). [CrossRef]
