Big data and AI-empowered classification of IR spectra acquired with a QCL-based standoff spectrometer: applications in forensics and security

We use a novel standoff IR spectrometer prototype to generate diffuse reflectance spectra of samples and evaluate the system’s performance for substance classification. Depending on the intended application, we use single- or multiplexed broadband External Cavity Quantum Cascade Lasers (EC-QCLs) as light source to probe the mid-infrared vibrational modes of samples with a measurement rate of one single- or multi-QCL-core spectrum per millisecond. The extremely short measurement times enable the generation of IR diffuse reflectance data with a rate that significantly overpasses the one achievable with FTIR spectrometers. This scales up the spectral information content that can be extracted from reference samples, so that substance-characteristic features are much better mirrored by the resulting classification models, especially for anisotropic samples. We test different big data modeling approaches for the acquired IR diffuse reflectance spectra, including PCA, OPLS-DA and machine-learning. We discuss classification results for a variety of samples including explosives (powder, paste, sheets and liquids), drugs (powders), and bodily fluids, including the ability to distinguish between animal and human blood as well as blood aging. We discuss a model for material classification between detonating cords and common electric cables and its potential application for the analysis of samples in the aftermath of explosions scenes. The presented results shall find their application in the development of more resilient, more selective and portable spectrometers for standoff substance detection and identification.