Algorithm-aided performance enhancement of a trace explosives sensor

Abstract

This article describes the exploitation of data obtained from a mature trace explosives detection device with the goal of enhancing the device’s performance via signal processing and no hardware modification. This is achieved by implementing a machine-learning algorithm based on nearest-neighbor binary classification. This study aims to determine the parameters of the algorithm defining key feature vectors for both explosive and non-explosive materials via systematic experimentation and measurements. Receiver operating characteristic (ROC) curves are estimated showing the trade-off between detection/false alarm rates, and the results demonstrate the merit of this approach for advancing the performance of this technology. Furthermore, the algorithm is shown to enable not just improved detection, but also a capability for target or materials identification.