Range-adaptive standoff recognition of explosive fingerprints on solid surfaces using a supervised learning method and laser-induced breakdown spectroscopy
I. Gaona, J. Serrano, J. Moros, J.J. Laserna, Analytical Chemistry, 2014, 86, 5045 – 5052
The distance between the sensor and the target is a particularly critical factor for an issue as crucial as explosive residues recognition when a laser-assisted spectroscopic technique operates in a standoff configuration. Particularly for laser ablation, variations in operational range influence the induced plasmas as well as the sensitivity of their ensuing optical emissions, thereby confining the attributes used in sorting methods. Though efficient classification models based on optical emissions gathered under specific conditions have been developed, their successful performance on any variable information is limited. Hence, to test new information by a designed model, data must be acquired under operational conditions totally matching those used during modeling. Otherwise, the new expected scenario needs to be previously modeled. To facing both this restriction and this time-consuming mission, a novel strategy is proposed in this work. On the basis of machine learning methods, the strategy stems from a decision boundary function designed for a defined set of experimental conditions. Next, particular semisupervised models to the envisaged conditions obtained adaptively on the basis of changes in laser fluence and light emission with variation of the sensor-to-target distance. Hence, the strategy requires only a little prior information, therefore ruling out the tedious and timeconsuming process of modeling all the expected distant scenes. Residues of ordinary materials (olive oil, fuel oil, motor oils, gasoline, car wax and hand cream) hardly cause confusion in alerting the presence of an explosive (DNT, TNT, RDX, or PETN) when tested within a range from 30 to SO m with varying laser irradiance between 8.2 and 1.3 GW cm(-2). With error rates of around 596, the experimental assessments confirm that this semisupervised model suitably addresses the recognition of organic residues on aluminum surfaces under different operational conditions.