# Mid-Infrared Laser Spectroscopy Detection and Quantification of Explosives in Soils Using Multivariate Analysis and Artificial Intelligence

^{1}

^{2}

^{3}

^{4}

^{5}

^{6}

^{7}

^{*}

## Abstract

**:**

## 1. Introduction

^{−1}. This spectral window corresponds to the fingerprint region of the HE samples deposited on painted metal car doors. Other studies have demonstrated diverse QCL approaches to the rapid identification and characterization of HEs, given the routine requirements of security checks [40,41].

## 2. Materials and Methods

#### 2.1. Reagents

#### 2.2. Sample Preparation

#### 2.3. Soil Characterization

#### 2.4. Data Acquisition and QCL System

^{−1}, 1111 to 1178 cm

^{−1}, and 1178 to 1600 cm

^{−1}The spectral linewidth was < 2 cm

^{−1}and the scan time was approximately 1.5 s for each of the diodes. The average power typically varied between 0.5 and 10 mW across the entire tuning range of ≈ 600 cm

^{−1}with 100:1 Transverse Electromagnetic Mode (TEM

_{00}) polarization and a beam divergence of < 2.5 mrad in the x-axis and < 5 mrad in the y-axis. The spectrometer had a 7.6-cm diameter ZnSe lens, which was used to focus the MIR beam to collect the reflected light and focus the light onto a thermoelectrically cooled mercury–cadmium–telluride (MCT) detector. The wavelength accuracy and precision were 0.5 and 0.2 cm

^{−1}, respectively. The spectroscopic system worked best at a distance to the target of 15 ± 3 cm, with each laser producing an elliptical spot with diameters of 4 and 2 mm in the same space at a distance of 15 cm due to the difference of beam divergence in the axes (Galan-Freyle et al. [30]).

#### 2.5. Data Quantification Analysis

^{2}), which indicates the percentage of variance present in the true component values reproduced by the PLS regression model. In contrast, the sensitivity (SEN) of multivariate methods can be estimated as the net analyte signal (NAS) [57] at a unit concentration as follows [58]:

**b**‖,

**b**denotes the regression vector of the PLS model. The analytical sensitivity (

**γ**) can be defined similarly to univariate calibrations [59] as

_{o})

^{1/2},

_{o}denotes the distance of the predicted sample to the mean of the calibration set at zero concentration and Δ(α,β) is a statistical parameter correlated with the α and β probabilities of falsely stating the presence or absence of the analyte. Δ(α,β) = 3.3 was used to compute the LOD values providing that the value for the degrees of freedom was > 25.

^{2})/(N − 1))

^{1/2},

#### 2.6. Pattern Recognition Analysis

#### 2.7. Artificial Intelligence Scheme

## 3. Results and Discussion

#### 3.1. Quantification Analysis

^{−1}). In this study, VN was used as a preprocessing step as it proved to be better than the other tested preprocessing steps except for KBr. Other preprocessing steps were tested, such as mean centering (MC), linear offset subtraction, straight-line subtraction, minimum and maximum normalization, multiplicative scatter correction, first derivative, second derivative, and no preprocessing step. When applying VN, the average intensity is calculated first, and then this value is subtracted from the spectrum. Next, the sum of the squared intensities is calculated, and the spectrum is divided by the square root of this sum. Models for DNT in KBr were generated to evaluate the detection in the absence of interferences: one at high concentrations (0–20%) and another at low concentrations (0–3%). The errors for these models are listed in Table 1. The most effective preprocessing method for the KBr models was MC. This result suggests that VN is a suitable preprocessing step only when interferences from the matrix are present. In a model free from the interfering matrix (KBr), applying VN for generating samples with low concentrations had the same signal intensity as samples with high concentrations. This is reflected in the lousy prediction and high uncertainty for samples with low concentrations (see Supplementary Materials: Figure S3).

#### Figures of Merit (FM)

**b**, which, in turn, were derived from the spectra and their respective concentrations of standards. VN directly affects the magnitude ‖

**b**‖ and value of SEN as a consequence. However, a better parameter for sensitivity can be obtained by calculating γ because this parameter is only affected by instrumental noises. The noise level was measured by collecting 20 spectra of a blank (target) and calculating an average of the standard deviations for all wavenumbers. The resulting noise for the models was different when considering 20 normalized spectra with VN (see Table 3). Otherwise, the γ-values calculated from the two types of models were very similar, which is an indication that γ is not affected by VN preprocessing or any other types of preprocessing.

^{−1}) provides an estimation of the minimum concentration difference (resultant) discernible by the model considering the instrumental noise as the only source of error. In the case of NAT-S Low, γ was 0.003% for the model with VN preprocessing and 0.002% for the model without preprocessing, with the difference being statistically insignificant. It is not possible to make a comparison of the sensitivities between the modes with matrix and without matrix because the magnitude of ‖

**b**‖ depends on the number of signals and number of latent variables (LV) (see Supplementary Materials: Figure S5). For the models of spectra with many signals (BC, SYN-S, and NAT-S), the magnitude of ‖

**b**‖ is higher than that for the models with low-intensity signals (KBr models). A better FM for this comparison is the LOD. For the models with the matrix (BC, SYN-S, and NAT-S), the LOD values are close to that of the model without the matrix (KBr models). Curves for the samples of low concentrations of DNT in NAT-S were generated to determine whether the employed concentration range influenced the LOD values of the curves. In these cases, the LOD decreased from 0.8% to 0.3%. DNT concentrations between 0.3% and 0.8% in the NAT-S Low model can be quantified with higher uncertainties and higher probabilities of false positives and missed detections because RSD should be between 10% and 33%.

^{2}(100 points) for each sample was generated. Two samples comprised the same soil as that used in the NAT-S models and were contaminated with 10% of DNT. The first sample consisted of a simple mixture (NAT-S-M). The second sample involved mixing and macerating the components (NAT-S-MM). The concentrations were predicted using the NAT-S model with three preprocessing methods: VN, MC, and no preprocessing. The map for the NAT-S model with VN is included in Supplementary Materials: Figures S6–S9, while the predictions for the 100 points for each map using different preprocessing methods are shown in Figure 4a,b. VN preprocessing applied to both samples (NAT-S-M and NAT-S-MM) resulted in better results, with the predicted values being close to the true value on average (10% DNT, see Figure 4c). MC preprocessing worked better in NAT-S-MM than NAT-S-M. This indicates that while the macerated process homogenized the size of the particles, MC was not able to compensate for the difference in the particle size. In contrast, the VN preprocessing method was able to compensate for these differences. The prediction of DNT in NAT-S-M shows peaks of high DNT concentrations (>10% DNT). This is because the particles of DNT were not homogenized. The models with no preprocessing provided bad predictions due to the difference in the baseline of the spectra. The above discussion demonstrates that VN preprocessing corrects the spectral variation due to changes in the particle size. To demonstrate the change in spectrum with the particle size, the tested soil was sieving for three particle sizes (d): d > 0.85 mm, 0.85 mm > d > 0.25 mm, and d < 0.25 mm. One hundred spectra were acquired at various locations on the sample surface for each value of d, and the average and standard deviation of the spectra were determined (see Supplementary Materials: Figures S10 and S11). The background offset spectral decreases with d, whereas the standard deviation increases with d; however, this pattern is not consistent throughout the spectra. It is higher in the 1000–1200-cm

^{−1}and 1400–1600-cm

^{−1}regions. This can be explained by the fact that MC is not able to correct the background offset spectral completely in contrast to VN; VN is better because it scales the spectrum to unit vectors, whereas MC only changes the baseline.

^{2}(100 points) were generated with the %DNT predicted from the NAT-S model using VN preprocessing. The mappings are present in Supplementary Materials: Figure S12, while predictions for each point are illustrated in Figure 4d. It can be noticed from the figure that the predictions were lower than the true values. This indicates that the NAT-S model quantifies below the true value of 10% DNT; however, it is capable of predicting the existence of an explosive in a soil. This indicates that the technique should be used in known soil to have good quantification.

^{−1}. PETN and RDX are nitro aliphatic explosives. TNT is very similar to DNT but considered to be a more challenging interference. Predictions for these samples were generated using the calibration curves for DNT/NAT-S. While the predicted values should have been zero or close to zero because the samples did not contain DNT, the average predicted concentrations were 8.8% (BA), 3.1% (IBP), -8.0% (PETN), 2.3% (RDX), and 25.8% (TNT) (see Table 4). The objective was to measure the model’s capability of discriminating against these interferences.

#### 3.2. Pattern Recognition Analysis

## 4. Conclusions

## Supplementary Materials

## Author Contributions

## Funding

## Acknowledgments

## Conflicts of Interest

## References

- Frische, T.; Höper, H. Soil microbial parameters and luminescent bacteria assays as indicators for in situ bioremediation of TNT-contaminated soils. Chemosphere
**2003**, 50, 415–427. [Google Scholar] [CrossRef] - Correa-Torres, S.N.; Pacheco-Londono, L.C.; Espinosa-Fuentes, E.A.; Rodriguez, L.; Souto-Bachiller, F.A.; Hernandez-Rivera, S.P. TNT removal from culture media by three commonly available wild plants growing in the Caribbean. J. Environ. Monit.
**2012**, 14, 30–33. [Google Scholar] [CrossRef] [PubMed] - Hildenbrand, J.; Herbst, J.; Wöllenstein, J.; Lambrecht, A. Explosive detection using infrared laser spectroscopy. Proc. SPIE
**2009**, 7222, 72220B. [Google Scholar] - Narang, U.; Gauger, P.R.; Ligler, F.S. A Displacement Flow Immunosensor for Explosive Detection Using Microcapillaries. Anal. Chem.
**1997**, 69, 2779–2785. [Google Scholar] [CrossRef] - Hilmi, A.; Luong, J.H.T. Micromachined Electrophoresis Chips with Electrochemical Detectors for Analysis of Explosive Compounds in Soil and Groundwater. Environ. Sci. Technol.
**2000**, 34, 3046–3050. [Google Scholar] [CrossRef] - Kumar, S.; Venkatramaiah, N.; Patil, S. Fluoranthene Based Derivatives for Detection of Trace Explosive Nitroaromatics. J. Phys. Chem. C
**2013**, 117, 7236–7245. [Google Scholar] [CrossRef] - Sheremata, T.W.; Halasz, A.; Paquet, L.; Thiboutot, S.; Ampleman, G.; Hawari, J. The Fate of the Cyclic Nitramine Explosive RDX in Natural Soil. Environ. Sci. Technol.
**2001**, 35, 1037–1040. [Google Scholar] [CrossRef] - Larson, S.L.; Martin, W.A.; Escalon, B.L.; Thompson, M. Dissolution, Sorption, and Kinetics Involved in Systems Containing Explosives, Water, and Soil. Environ. Sci. Technol.
**2008**, 42, 786–792. [Google Scholar] [CrossRef] - Marple, R.L.; LaCourse, W.R. Application of Photoassisted Electrochemical Detection to Explosive-Containing Environmental Samples. Anal. Chem.
**2005**, 77, 6709–6714. [Google Scholar] [CrossRef] - Gallagher, N.B.; Blake, T.A.; Gassman, P.L. Application of extended inverse scatter correction to mid-infrared reflectance spectra of soil. J. Chemom.
**2005**, 19, 271–281. [Google Scholar] [CrossRef] - Forouzangohar, M.; Kookana, R.S.; Forrester, S.T.; Smernik, R.J.; Chittleborough, D.J. Mid-infrared Spectroscopy and Chemometrics to Predict Diuron Sorption Coefficients in Soils. Environ. Sci. Technol.
**2008**, 42, 3283–3288. [Google Scholar] [CrossRef] - Gallagher, N.B.; Gassman, P.L.; Blake, T.A. Strategies for Detecting Organic Liquids on Soils Using Mid-Infrared Reflection Spectroscopy. Environ. Sci. Technol.
**2008**, 42, 5700–5705. [Google Scholar] [CrossRef] [PubMed] - Mukherjee, A.; Von der Porten, S.; Patel, C.K.N. Standoff detection of explosive substances at distances of up to 150 m. Appl. Opt.
**2010**, 49, 2072–2078. [Google Scholar] [CrossRef] - Hernández, M.D.; Santiago, I.; Padilla, I.Y. Macro-sorption of 2,4-dinitrotoluene onto sandy and clay soils. Proc. SPIE
**2006**, 6217, 621736. [Google Scholar] - Baez, B.; Correa, S.N.; Hernandez-Rivera, S.P.; de Jesus, M.; Castro, M.E.; Mina, N.; Briano, J.G. Transport of explosives I: TNT in soil and its equilibrium vapor. Proc. SPIE
**2004**, 5415, 1389–1399. [Google Scholar] - Torres, A.; Padilla, I.; Hwang, S. Physical modeling of 2,4-DNT gaseous diffusion through unsaturated soil. Proc. SPIE
**2007**, 6553, 65531Q. [Google Scholar] - Herrera-Sandoval, G.M.; Ballesteros, L.M.; Mina, N.; Briano, J.; Castro, M.E.; Hernandez-Rivera, S.P. Raman signatures of TNT in contact with sand particles. Proc. SPIE
**2005**, 5794, 1245–1253. [Google Scholar] - Blanco, A.; Mina, N.; Castro, M.E.; Castillo-Chara, J.; Hernandez-Rivera, S.P. Effect of environmental conditions on the spectroscopic signature of DNT in sand. Proc. SPIE
**2005**, 5794, 1281–1289. [Google Scholar] - Ballesteros, L.M.; Herrera, G.M.; Castro, M.E.; Briano, J.; Mina, N.; Hernandez-Rivera, S.P. Spectroscopic signatures of PETN in contact with sand particles. Proc. SPIE
**2005**, 5794, 1254–1262. [Google Scholar] - Hernandez-Rivera, S.P.; Manrique-Bastidas, C.A.; Blanco, A.; Primera, O.M.; Pacheco, L.C.; Castillo-Chara, J.; Castro, M.E.; Mina, N. Spectroscopic characterization of nitroaromatic landmine signature explosives. Proc. SPIE
**2004**, 5415, 474–485. [Google Scholar] - Osorio, C.; Gomez, L.M.; Hernandez, S.P.; Castro, M.E. Time-of-flight mass spectroscopy measurements of TNT and RDX on soil surfaces. Proc. SPIE
**2005**, 5794, 803–811. [Google Scholar] - Manrique-Bastidas, C.A.; Mina, N.; Castro, M.E.; Hernandez-Rivera, S.P. Raman microspectroscopy and FTIR crystallization studies of 2,4,6-TNT in soil. Proc. SPIE
**2005**, 5794, 1358–1365. [Google Scholar] - Blanco, A.; Pacheco-Londoño, L.C.; Peña-Quevedo, A.J.; Hernández-Rivera, S.P. UV Raman detection of 2,4-DNT in contact with sand particles. Proc. SPIE
**2006**, 6217, 621737. [Google Scholar] - Galán-Freyle, N.J.; Pacheco-Londoño, L.C.; Figueroa-Navedo, A.M.; Hernandez-Rivera, S.P. Standoff Detection of Highly Energetic Materials Using Laser-Induced Thermal Excitation of Infrared Emission. Appl. Spectrosc.
**2015**, 69, 535–544. [Google Scholar] [CrossRef] [PubMed] - Galán-Freyle, N.J.; Ospina-Castro, M.L.; Medina-González, A.R.; Villarreal-González, R.; Hernández-Rivera, S.P.; Pacheco-Londoño, L.C. Artificial Intelligence Assisted Mid-Infrared Laser Spectroscopy In Situ Detection of Petroleum in Soils. Appl. Sci.
**2020**, 10, 1319. [Google Scholar] [CrossRef][Green Version] - Rüther, A.; Pfeifer, M.; Lórenz-Fonfría, V.A.; Lüdeke, S. pH Titration Monitored by Quantum Cascade Laser-Based Vibrational Circular Dichroism. J. Phys. Chem. B
**2014**, 118, 3941–3949. [Google Scholar] [CrossRef] - Lüdeke, S.; Pfeifer, M.; Fischer, P. Quantum-Cascade Laser-Based Vibrational Circular Dichroism. J. Am. Chem. Soc.
**2011**, 133, 5704–5707. [Google Scholar] [CrossRef] - Shi, Q.; Nelson, D.D.; McManus, J.B.; Zahniser, M.S.; Parrish, M.E.; Baren, R.E.; Shafer, K.H.; Harward, C.N. Quantum Cascade Infrared Laser Spectroscopy for Real-Time Cigarette Smoke Analysis. Anal. Chem.
**2003**, 75, 5180–5190. [Google Scholar] [CrossRef] - Wörle, K.; Seichter, F.; Wilk, A.; Armacost, C.; Day, T.; Godejohann, M.; Wachter, U.; Vogt, J.; Radermacher, P.; Mizaikoff, B. Breath Analysis with Broadly Tunable Quantum Cascade Lasers. Anal. Chem.
**2013**, 85, 2697–2702. [Google Scholar] [CrossRef] - Galán-Freyle, N.J.; Pacheco-Londoño, L.C.; Román-Ospino, A.D.; Hernandez-Rivera, S.P. Applications of Quantum Cascade Laser Spectroscopy in the Analysis of Pharmaceutical Formulations. Appl. Spectrosc.
**2016**, 70, 1511–1519. [Google Scholar] [CrossRef] - Padilla-Jiménez, A.C.; Ortiz-Rivera, W.; Rios-Velazquez, C.; Vazquez-Ayala, I.; Hernández-Rivera, S.P. Detection and discrimination of microorganisms on various substrates with quantum cascade laser spectroscopy. OPTICE
**2014**, 53, 061611. [Google Scholar] [CrossRef] - Faist, J.; Capasso, F.; Sivco, D.L.; Sirtori, C.; Hutchinson, A.L.; Cho, A.Y. Quantum Cascade Laser. Science
**1994**, 264, 553–556. [Google Scholar] [CrossRef] [PubMed] - Hvozdara, L.; Pennington, N.; Kraft, M.; Karlowatz, M.; Mizaikoff, B. Quantum cascade lasers for mid-infrared spectroscopy. Vib. Spectrosc.
**2002**, 30, 53–58. [Google Scholar] [CrossRef] - Ruiz-Caballero, J.L.; Blanco-Riveiro, L.A.; Ramirez-Marrero, I.A.; Perez-Almodovar, L.A.; Colon-Mercado, A.M.; Castro-Suarez, J.R.; Pacheco-Londoño, L.C.; Hernandez-Rivera, S.P. Enhanced RDX Detection Studies on Various Types of Substrates via Tunable Quantum Cascade Laser Spectrometer Coupled with Grazing Angle Probe. IOP Conf. Ser. Mater. Sci. Eng.
**2019**, 519, 012007. [Google Scholar] [CrossRef] - Pacheco-Londoño, L.C.; Galán-Freyle, N.J.; Figueroa-Navedo, A.M.; Infante-Castillo, R.; Ruiz-Caballero, J.L.; Hernández-Rivera, S.P. Quantum cascade laser back-reflection spectroscopy at grazing-angle incidence using the fast Fourier transform as a data preprocessing algorithm. J. Chemom.
**2019**, 33, e3167. [Google Scholar] [CrossRef] - Pacheco-Londoño, L.C.; Aparicio-Bolaño, J.A.; Galán-Freyle, N.J.; Román-Ospino, A.D.; Ruiz-Caballero, J.L.; Hernández-Rivera, S.P. Classical Least Squares-Assisted Mid-Infrared (MIR) Laser Spectroscopy Detection of High Explosives on Fabrics. Appl. Spectrosc.
**2019**, 73, 17–29. [Google Scholar] [CrossRef] - Pacheco-Londoño, L.C.; Ruiz-Caballero, J.L.; Ramírez-Cedeño, M.L.; Infante-Castillo, R.; Gálan-Freyle, N.J.; Hernández-Rivera, S.P. Surface Persistence of Trace Level Deposits of Highly Energetic Materials. Molecules
**2019**, 24, 3494. [Google Scholar] [CrossRef][Green Version] - Pacheco-Londoño, L.C.; Castro-Suarez, J.R.; Galán-Freyle, N.J.; Figueroa-Navedo, A.M.; Ruiz-Caballero, J.L.; Infante-Castillo, R.; Hernández-Rivera, S.P. Mid-Infrared Laser Spectroscopy Applications I: Detection of Traces of High Explosives on Reflective and Matte Substrates. In Infrared Spectroscopy—Principles, Advances, and Applications; IntechOpen: London, UK, 2019. [Google Scholar] [CrossRef][Green Version]
- Phillips, M.C.; Bernacki, B.E. Hyperspectral microscopy of explosives particles using an external cavity quantum cascade laser. OPTICE
**2012**, 52, 061302. [Google Scholar] [CrossRef][Green Version] - Pacheco-Londoño, L.C.; Castro-Suarez, J.R.; Hernández-Rivera, S.P. Detection of Nitroaromatic and Peroxide Explosives in Air Using Infrared Spectroscopy: QCL and FTIR. Adv. Opt. Technol.
**2013**, 2013, 532670. [Google Scholar] [CrossRef][Green Version] - Sirkeli, V.P.; Yilmazoglu, O.; Preu, S.; Küppers, F.; Hartnagel, H.L. Proposal for a Monolithic Broadband Terahertz Quantum Cascade Laser Array Tailored to Detection of Explosive Materials. Sens. Lett.
**2018**, 16, 1–7. [Google Scholar] [CrossRef] - Pettersson, A.; Wallin, S.; Östmark, H.; Ehlerding, A.; Johansson, I.; Nordberg, M.; Ellis, H.; Al-Khalili, A. Explosives standoff detection using Raman spectroscopy: From bulk towards trace detection. Proc. SPIE
**2010**, 7664, 76641K. [Google Scholar] - Yang, C.S.C.; Brown, E.E.; Hommerich, U.; Jin, F.; Trivedi, S.B.; Samuels, A.C.; Snyder, A.P. Long-Wave, Infrared Laser-Induced Breakdown (LIBS) Spectroscopy Emissions from Energetic Materials. Appl. Spectrosc.
**2012**, 66, 1397–1402. [Google Scholar] [CrossRef] - Misra, A.K.; Sharma, S.K.; Acosta, T.E.; Porter, J.N.; Bates, D.E. Single-Pulse Standoff Raman Detection of Chemicals from 120 m Distance During Daytime. Appl. Spectrosc.
**2012**, 66, 1279–1285. [Google Scholar] [CrossRef] - Gottfried, J.L.; De Lucia, F.C.; Munson, C.A.; Miziolek, A.W. Standoff Detection of Chemical and Biological Threats Using Laser-Induced Breakdown Spectroscopy. Appl. Spectrosc.
**2008**, 62, 353–363. [Google Scholar] [CrossRef] [PubMed] - Castro-Suarez, J.R.; Pacheco-Londoño, L.C.; Vélez-Reyes, M.; Diem, M.; Tague, T.J.; Hernandez-Rivera, S.P. FT-IR Standoff Detection of Thermally Excited Emissions of Trinitrotoluene (TNT) Deposited on Aluminum Substrates. Appl. Spectrosc.
**2013**, 67, 181–186. [Google Scholar] [CrossRef] [PubMed] - Carter, J.C.; Angel, S.M.; Lawrence-Snyder, M.; Scaffidi, J.; Whipple, R.E.; Reynolds, J.G. Standoff Detection of High Explosive Materials at 50 Meters in Ambient Light Conditions Using a Small Raman Instrument. Appl. Spectrosc.
**2005**, 59, 769–775. [Google Scholar] [CrossRef] [PubMed] - Averett, L.A.; Griffiths, P.R. Mid-Infrared Diffuse Reflection of a Strongly Absorbing Analyte on Non-Absorbing and Absorbing Matrices. Part II: Thin Liquid Layers on Powdered Substrates. Appl. Spectrosc.
**2008**, 62, 383–388. [Google Scholar] [CrossRef] - Pacheco-Londoño, L.; Ortiz-Rivera, W.; Primera-Pedrozo, O.; Hernández-Rivera, S. Vibrational spectroscopy standoff detection of explosives. Anal. Bioanal. Chem.
**2009**, 395, 323–335. [Google Scholar] [CrossRef] - Van Neste, C.W.; Senesac, L.R.; Thundat, T. Standoff Spectroscopy of Surface Adsorbed Chemicals. Anal. Chem.
**2009**, 81, 1952–1956. [Google Scholar] [CrossRef] - Moros, J.; Lorenzo, J.A.; Lucena, P.; Miguel Tobaria, L.; Laserna, J.J. Simultaneous Raman Spectroscopy−Laser-Induced Breakdown Spectroscopy for Instant Standoff Analysis of Explosives Using a Mobile Integrated Sensor Platform. Anal. Chem.
**2010**, 82, 1389–1400. [Google Scholar] [CrossRef] - Moros, J.; Laserna, J.J. New Raman–Laser-Induced Breakdown Spectroscopy Identity of Explosives Using Parametric Data Fusion on an Integrated Sensing Platform. Anal. Chem.
**2011**, 83, 6275–6285. [Google Scholar] [CrossRef] [PubMed] - Ortiz-Rivera, W.; Pacheco-Londoño, L.; Castro-Suarez, J.; Felix-Rivera, H.; Hernandez-Rivera, S. Vibrational Spectroscopy Standoff Detection of Threat Chemicals; SPIE: Bellingham, WA, USA, 2011; Volume 8031. [Google Scholar]
- Miyazawa, M.; Pavan, M.A.; de Oliveira, E.L.; Ionashiro, M.; Silva, A.K. Gravimetric determination of soil organic matter. Braz. Arch. Biol. Technol.
**2000**, 43, 475–478. [Google Scholar] [CrossRef] - Weiner, E.R. Applications of Environmental Chemistry: A Practical Guide for Environmental Professionals; Lewis Pub.: Boca Raton, FL, USA, 2000. [Google Scholar]
- Yanjun, C.; Achari, G.; Langford, C.H. Protocols for the analysis of transformer oil and its degradation in soil by hydrogen peroxide. Can. J. Civ. Eng.
**2009**, 36, 1547–1557. [Google Scholar] [CrossRef] - Lorber, A. Error propagation and figures of merit for quantification by solving matrix equations. Anal. Chem.
**1986**, 58, 1167–1172. [Google Scholar] [CrossRef] - Ferreira, M.H.; Braga, J.W.B.; Sena, M.M. Development and validation of a chemometric method for direct determination of hydrochlorothiazide in pharmaceutical samples by diffuse reflectance near-infrared spectroscopy. Microelectron. J.
**2013**, 109, 158–164. [Google Scholar] [CrossRef] - Olivieri, A.C.; Faber, N.M.; Ferré, J.; Boqué, R.; Kalivas, J.H.; Mark, H. Uncertainty estimation and figures of merit for multivariate calibration (IUPAC Technical Report). Pure Appl. Chem.
**2006**, 78, 633. [Google Scholar] [CrossRef][Green Version] - Felipe-Sotelo, M.; Cal-Prieto, M.J.; Ferre, J.; Boque, R.; Andrade, J.M.; Carlosena, A. Linear PLS regression to cope with interferences of major concomitants in the determination of antimony by ETAAS. J. Anal. Spectrom.
**2006**, 21, 61–68. [Google Scholar] [CrossRef] - Galan-Freyle, N.J.; Figueroa-Navedo, A.M.; Pacheco-Londoño, Y.C.; Ortiz-Rivera, W.; Pacheco-Londoño, L.C.; Hernández-Rivera, S.P. Chemometrics-enhanced fiber-optic Raman detection, discrimination and quantification of chemical agents simulants concealed in commercial bottles. Anal. Chem. Res.
**2014**, 2, 15–22. [Google Scholar] [CrossRef][Green Version] - Buitinck, L.; Louppe, G.; Blondel, M.; Pedregosa, F.; Mueller, A.; Grisel, O.; Niculae, V.; Prettenhofer, P.; Gramfort, A.; Grobler, J.; et al. API design for machine learning software: Experiences from the scikit-learn project. In Proceedings of the European Conference on Machine Learning and Principles and Practices of Knowledge Discovery in Databases, Prague, Czech Republic, 23 September 2013. [Google Scholar]
- Kubelka, P. New Contributions to the Optics of Intensely Light-Scattering Materials. Part I. J. Opt. Soc. Am
**1948**, 38, 448. [Google Scholar] [CrossRef] - Bull, C.R. Compensation for particle size effects in near-infrared reflectance. Analyst
**1991**, 116, 781–786. [Google Scholar] [CrossRef] - Sirita, J.; Phanichphant, S.; Meunier, F.C. Quantitative Analysis of Adsorbate Concentrations by Diffuse Reflectance FT-IR. Anal. Chem.
**2007**, 79, 3912–3918. [Google Scholar] [CrossRef] [PubMed][Green Version] - Igne, B.; Hurburgh, C.R. Local chemometrics for samples and variables: Optimizing calibration and standardization processes. J. Chemom.
**2010**, 24, 75–86. [Google Scholar] [CrossRef]

**Figure 2.**Predicted values vs. real values of % DNT from partial least squares (PLS) models on the matrices: (

**a**) bentonite clay (BC); (

**b**) synthetic soil (SYN-S); (

**c**) natural soil (NAT-S); (

**d**) NAT-S to low concentration (NAT-S Low); (

**e**) potassium bromide (KBr) and (

**f**) KBr to low concentration (KBr Low).

**Figure 3.**DNT spectrum in the investigated soil matrices in K-M (K-M for T: K-M calculate from transmittance for red spectra): (

**a**) BC; (

**b**) SYN-S; (

**c**) NAT-S, and (

**d**) NAT-S Low.

**Figure 4.**Predictions for 100 pts. maps after applying VN, mean centering (MC), and no preprocessing: (

**a**) simple mixture (NAT-S-M); (

**b**) macerating the components (NAT-S-MM); (

**c**) Averages for the predictions of NAT-S-M and NAT-S-MM (

**d**) NAT2-S with VN preprocessing.

**Figure 5.**Dependence of the root-mean-square error of cross-validation (RMSECV) on LV for the considered interferences: the NAT-S optimization of the validation set (OPT-Val) and NAT-S models.

**Figure 6.**(

**a**) Probability distribution of the sum of ${\beta}_{j}$ (SUM (β)) for a simple binary discrimination model. (

**b**) Comparison of the tested machine learning (ML) methods in terms of their log-loss and accuracy.

**Table 1.**Accuracy, bias, and the number of latent variables (LV) for the studied models and matrices.

RMSEE | RMSECV | RMSEP | Bias | LV | |
---|---|---|---|---|---|

BC | 0.41 | 0.57 | 0.70 | −0.0007 | 11 |

SYN-S | 0.35 | 0.43 | 0.53 | −0.0069 | 7 |

NAT-S | 0.25 | 0.39 | 0.39 | 0.0010 | 10 |

NAT-S Low | 0.08 | 0.10 | 0.34 | 0.0100 | 7 |

KBr | 0.18 | 0.32 | 0.41 | −0.0044 | 10 |

KBr Low | 0.02 | 0.03 | 0.08 | −0.0003 | 9 |

**Table 2.**Relative standard deviation (RSD) and relative predictive determinant (RPD) values for the models generated.

RSDr | RSDh | RSDrd | RPD-CV | RPD-Test | |
---|---|---|---|---|---|

BC | 2.4 | 5.5 | 12.7 | 9.7 | 11.3 |

SYN-S | 1.5 | 5.3 | 9.1 | 9.2 | 16.2 |

NAT-S | 2.2 | 4.2 | 4.8 | 16.1 | 27.6 |

NAT-S Low | 3.2 | 13.8 | 32.9 | 9.1 | 12.6 |

KBr | 3.4 | 3.6 | 4.1 | 19.8 | 36.5 |

KBr Low | 5.7 | 6.2 | 11.9 | 27.5 | 50.1 |

**Table 3.**Sensitivity, analytical sensitivity, the limit of detection (LOD), and noise of the models, with and without vector normalization (VN) preprocessing.

VN | ||||

LOD | SEN | γ | γ^{−1} | |

BC | 1.4 | 0.013 | 43 | 0.023 |

SYN-S | 1.2 | 0.016 | 53 | 0.019 |

NAT-S | 0.8 | 0.010 | 32 | 0.031 |

NAT-S Low | 0.3 | 0.087 | 289 | 0.003 |

Noise = 0.0003 | ||||

No Preprocessing | ||||

LOD | SEN | γ | γ^{−1} | |

BC | 2.3 | 1.48 | 39 | 0.025 |

SYN-S | 1.4 | 2.25 | 60 | 0.017 |

NAT-S | 1.0 | 1.36 | 36 | 0.028 |

NAT-S Low | 0.6 | 20.63 | 552 | 0.002 |

KBr | 0.6 | 0.51 | 14 | 0.072 |

KBr Low | 0.07 | 1.26 | 34 | 0.029 |

Noise = 0.037 |

NAT-S | NAT-S OPT-Val | |
---|---|---|

BA | 8.8 | 0.02 |

IBP | 3.8 | −0.02 |

PENT | −8.0 | 0.00 |

RDX | 2.3 | 0.02 |

TNT | 25.8 | 0.11 |

NAT-S | NAT-S OPT-Val | |
---|---|---|

R^{2} cal | 99.87 | 99.65 |

R^{2} val | 99.61 | 98.92 |

R^{2} test | 99.63 | 98.07 |

RMSEE | 0.245% | 0.425% |

RMSECV | 0.39% | 0.72% |

RMSEP | 0.39% | 0.88% |

Bias | 0.001 | −0.0016 |

LV | 10 | 19 |

ho | 0.06 | 0.021 |

LOD | 0.80% | 1.4% |

RPD | 16.1 | 9.61 |

**Table 6.**Confusion matrix and evaluation criteria for the binary discrimination model using SUM (β) parameters.

Precision | Recall | F1-Score | Support | Accuracy | Matrix of Confusion | ||
---|---|---|---|---|---|---|---|

Model | EXP | NONE | |||||

EXP | 0.768 | 1.000 | 0.869 | 1126 | 0.877 | 865 | 261 |

NONE | 1.000 | 0.793 | 0.884 | 997 | 0 | 997 |

Precision | Recall | f1-Score | Support | Accuracy | Confusion Matrix | ||||
---|---|---|---|---|---|---|---|---|---|

Model | DNT | NONE | RDX | TNT | |||||

DNT | 0.997 | 0.997 | 0.997 | 347 | 0.997 | 346 | 1 | 0 | 0 |

NONE | 1.000 | 0.996 | 0.998 | 564 | 0 | 564 | 0 | 0 | |

RDX | 0.993 | 1.000 | 0.997 | 149 | 1 | 0 | 148 | 0 | |

TNT | 0.990 | 1.000 | 0.995 | 100 | 0 | 1 | 0 | 99 | |

Test | DNT | NONE | RDX | TNT | |||||

DNT | 1.000 | 0.997 | 0.998 | 289 | 0.996 | 289 | 0 | 0 | 0 |

NONE | 0.993 | 1.000 | 0.997 | 433 | 0 | 430 | 1 | 2 | |

RDX | 0.993 | 0.993 | 0.993 | 139 | 1 | 0 | 138 | 0 | |

TNT | 1.000 | 0.981 | 0.990 | 102 | 0 | 0 | 0 | 102 |

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Pacheco-Londoño, L.C.; Warren, E.; Galán-Freyle, N.J.; Villarreal-González, R.; Aparicio-Bolaño, J.A.; Ospina-Castro, M.L.; Shih, W.-C.; Hernández-Rivera, S.P. Mid-Infrared Laser Spectroscopy Detection and Quantification of Explosives in Soils Using Multivariate Analysis and Artificial Intelligence. *Appl. Sci.* **2020**, *10*, 4178.
https://doi.org/10.3390/app10124178

**AMA Style**

Pacheco-Londoño LC, Warren E, Galán-Freyle NJ, Villarreal-González R, Aparicio-Bolaño JA, Ospina-Castro ML, Shih W-C, Hernández-Rivera SP. Mid-Infrared Laser Spectroscopy Detection and Quantification of Explosives in Soils Using Multivariate Analysis and Artificial Intelligence. *Applied Sciences*. 2020; 10(12):4178.
https://doi.org/10.3390/app10124178

**Chicago/Turabian Style**

Pacheco-Londoño, Leonardo C., Eric Warren, Nataly J. Galán-Freyle, Reynaldo Villarreal-González, Joaquín A. Aparicio-Bolaño, María L. Ospina-Castro, Wei-Chuan Shih, and Samuel P. Hernández-Rivera. 2020. "Mid-Infrared Laser Spectroscopy Detection and Quantification of Explosives in Soils Using Multivariate Analysis and Artificial Intelligence" *Applied Sciences* 10, no. 12: 4178.
https://doi.org/10.3390/app10124178