Multivariate Discrimination Model for TNT and Gunpowder Using an Electronic Nose Prototype: A Proof of Concept

Abstract

In this work a proof of concept for discriminating explosive substances is presented, where a discrimination model for the classification of TNT and gunpowder is developed. An electronic nose was designed for sensing volatile organic compounds present in TNT and gunpowder, and a model that combines Principal Component Analysis and Fisher Discriminant Analysis was built for enhancing class discrimination. The model was tested in two scenarios: discriminating among the two explosive substances and one non-explosive, and discriminating between explosives and non-explosives, obtaining better results in the second case. In order to test model confidence a permutation test was used proving an accuracy of 67% with a p-value <0.01 for the first scenario, and an accuracy of 86.6% for the second one. These results make us think that by enhancing the prototype characteristics in both hardware and software, we would be able to achieve better results.