Fast standoff investigation of chemical and biological samples using laser induced fluorescence signals, machine learning and an interactive interface

Abstract

Release of hazardous substances may cause severe consequences to humans and infrastructure. A fast detection system classifying these substances can be used to initiate countermeasures quickly and reduce damage to general public significantly. Nowadays, the investigation of such materials is time and money consuming but essential for damage limitation. Current procedures take long and require methods that involve measurements at close range and/or sampling for subsequent laboratory analyses. Both, sampling time and distance, can be improved using standoff laser induced fluorescence (LIF) spectroscopy which enables detection in seconds reaching distances over 100 m. By now the specificity of the technique is not sufficient to identify samples but it can be helpful for risk assessment and to guide first responders to salient regions for subsequent in situ measurements. This contribution presents an interactive graphical user interface as well as the practical workflow from generating training data and classification models to forecasting new records concurrently after the measurement. The foregoing modeling process is based on datasets generated previously with well-defined samples. Each sample from a considerable set of different chemical, botanical and bacterial substances can be distinguished using the LIF signals excited with short laser pulses of two UV wavelengths within a few seconds. Simultaneously, the fluorescence lifetime is recorded to provide additional information for a further enhanced discriminability. Estimating the sample species for new measurements consumes just a few seconds - including data acquisition, preprocessing, model application and visualization to the operator. As an example, the workflow is presented together with performance results for a test classification of 20 different substances.