Machine Learning Augmented Detection of Trace Explosives on Surfaces Using Active Longwave Infrared Imaging.

Abstract

There is a need for rapid non-contact detection of trace explosives on arbitrary surfaces for applications such as checkpoint screening and post-blast forensics. In this work, active longwave infrared imaging, using a tunable quantum cascade laser light source, is shown to be able to detect three common explosives on typical luggage material substrates at surface concentrations as low as 230 ng/cm². Use of a random forest machine learning algorithm, trained using synthetically produced spectra, allows performance beyond predicted limitations of speckle noise. The random forest algorithm further allows discrimination against ambient dust without specific training. Use of high collection angles reduces the dynamic range requirement of the system without impairing performance. Potential system improvements are discussed to meet the area coverage rate needs of several applications.