Evolutionary optimization of THz spectral data for the quantitative identification of contraband substances.

The identification of contraband materials poses a significant challenge in law enforcement, particularly due to legal restrictions prohibiting search without a warrant, and with commercial X-ray scanning, one cannot distinguish between a legal and an illegal drug when placed in dielectric packaging. Terahertz (THz) radiation, spanning frequencies from 0.1 to 10 THz, lying between the microwave and infrared regions of the electromagnetic spectrum, penetrates such packages as long as they are not metallic in nature. Several compounds have unique spectral signatures in the THz regime, enabling the identification of controlled substances, even in cases of complex mixtures. This present work integrates evolutionary algorithms and machine learning approaches to highlight the potential to automate the identification and quantification of several drug components within complex, real-world mixtures, advancing the application of THz spectroscopy. Using fundamental optical properties derived from the THz spectra, we focused on three representative compounds: caffeine and pseudo-scents of heroin and marijuana, with their respective absorbance, extinction coefficient, and absorption coefficient as the key indicators for the process. We trained a model to estimate drug concentrations by aggregating their spectral data with the evolutionary micro-genetic algorithm (μGA) and partial least square regression. With average errors around 4% for caffeine and 9% for pseudo-scents of heroin and marijuana, our predictions closely match the true concentrations, obtaining R2 values around 0.99. These degrees of accuracy imply the potential for the approach to be developed into a practical tool for non-invasive, reliable drug screening in sensitive environments like airports, mail facilities, and border control, integrated with a sensitive and portable THz device.