Real-time analysis of data provides input for decision makers. However, in the battlefield, that could be the difference between life and death. Therefore, techniques must be developed that provide data in a way that can be reduced to real-time information. Hyperspectral data is often sought after as it provides spatial-spectral information but comes with a large computation cost. Real-time analysis of hyperspectral data is often difficult after an appreciable amount of time due to the volume of data that must be analyzed. However, commercial off-the-shelf instrumentation that normally outputs large hypercubes of information can be computationally managed in a way such that real-time processing is achievable at low levels of analyte. In this work, we show near-trace level anomaly detection of explosive precursors, explosives, and pharmaceutical surrogates on real-world surfaces using a commercial off-the-shelf instrument. The threat anomaly detection (ThreAD) algorithm that is employed uses a semi-supervised machine learning method to determine where the anomalous data (i.e. analyte) is present. This work will provide approximate limits of anomaly detection (LOADs) for some analyte/surface combinations in laboratory conditions.
|