Recognition of explosives fingerprints on objects for courier services using machine learning methods and laser-induced breakdown spectroscopy J. Moros a , J. Serrano a , F.J. Gallego b , J. Macı ´as b , J.J. Laserna a,n a Department of Analytical Chemistry, University of Malaga, E-29071 Malaga, Spain b Department of Mathematical Analysis, University of Malaga, E-29071 Malaga, Spain article info Article history: Received 7 November 2012 Received in revised form 1 February 2013 Accepted 11 February 2013 Available online 19 February 2013 Keywords: LIBS Fingerprints Home-made explosives Harmless Machine learning Decision tree abstract During recent years laser-induced breakdown spectroscopy (LIBS) has been considered one of the techniques with larger ability for trace detection of explosives. However, despite of the high sensitivity exhibited for this application, LIBS suffers from a limited selectivity due to difficulties in assigning the molecular origin of the spectral emissions observed. This circumstance makes the recognition of fingerprints a latent challenging problem. In the present manuscript the sorting of six explosives (chloratite, ammonal, DNT, TNT, RDX and PETN) against a broad list of potential harmless interferents (butter, fuel oil, hand cream, olive oil, y), all of them in the form of fingerprints deposited on the surfaces of objects for courier services, has been carried out. When LIBS information is processed through a multi-stage architecture algorithm built from a suitable combination of 3 learning classifiers, an unknown fingerprint may be labeled into a particular class. Neural network classifiers trained by the Levenberg–Marquardt rule were decided within 3D scatter plots projected onto the subspace of the most useful features extracted from the LIBS spectra. Experimental results demonstrate that the presented algorithm sorts fingerprints according to their hazardous character, although its spectral information is virtually identical in appearance, with rates of false negatives and false positives not beyond of 10%. These reported achievements mean a step forward in the technology readiness level of LIBS for this complex application related to defense, homeland security and force protection. & 2013 Elsevier B.V. All rights reserved. 1. Introduction In recent times, laser-induced breakdown spectroscopy (LIBS) has been demonstrated to be a powerful analytical tool to cope the direct chemical detection of energetic materials and residues of explosives in real-time [1–3]. However, despite all the efforts and the gains achieved in recent years when using LIBS for this last topic [4,5], a particularly challenging problem still needs to be overcome for offering greater assurances in this matter. From the identification point of view, the recognition aspects of LIBS are considered one of its Achilles’ heels, because compounds sharing a similar chemical composition also have virtually the same LIBS signature. Difficulties on discrimination issues are increasing when residues and also the support where they are left are from the same nature. The situation is further complicated due to the fact that for heterogeneous residues their particular emissions and their characteristic intensity ratios, which are usually used in classification of explosives by LIBS, may fluctuate in a random manner. Finally, to round off the problem, the varying influence of the surrounding atmosphere may also contribute to these cited fluctuations. For all these reasons, a broad range of powerful chemometric tools have been applied for dealing with this issue of assigning an identity to a residue, or at least, for its sorting into as hazardous or harmless. Table 1 summarizes the state-of-the-art of statistical and mathematical tools applied to LIBS information in order to improve its discrimination capability. The most relevant aspects involved in the successful application of these multivariate analysis techni- ques, namely, the evaluated materials (not only explosives but also potential interferents), the supports where the residues are left, the way used to prepare the experimental targets, and supplied quantities, are provided. In any case, readers interested in detailed information are invited to consult the references cited. Among the multivariate techniques demonstrated to be viable to classify an unknown sample as an explosive or a harmless product, the most widely used has been the principal component analysis (PCA) [6–9]. This technique works well when the varia- bility within a group is much smaller than the variability among groups. However, this circumstance is unlikely when account is taken of the heterogeneous deposition of a fingerprint and the large shot-to-shot variability of LIBS spectra. For these reasons, the deposition of discrete volumes of acetone solutions and the Contents lists available at SciVerse ScienceDirect journal homepage: www.elsevier.com/locate/talanta Talanta 0039-9140/$ - see front matter & 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.talanta.2013.02.026 n Corresponding author. Tel.: þ34 952 13 1881; fax: þ34 952 13 2000. E-mail address: laserna@uma.es (J.J. Laserna). Talanta 110 (2013) 108–117