Wide dynamic range and real-time reagent identification and imaging using multi-wavelength terahertz parametric generation and machine learning

Abstract

In this study, we propose a technique for identifying and imaging reagents through shielding over a wide dynamic range using a real-time terahertz (THz) spectroscopy system with multi-wavelength THz parametric generation/detection and machine learning. To quickly identify reagents through shielding, the spectral information of the “detection Stokes beam” is used for reagent recognition via machine learning. In general THz wave-based reagent identification, continuous spectra are acquired and analyzed quantitatively by post-processing. In actual applications, however, such as testing for illicit drugs in mail, the technology must be able to quickly identify reagents as opposed to quantifying the amount present. In multi-wavelength THz parametric generation/detection, THz spectral information can be measured instantly using a “multi-wavelength detection Stokes beam” and near-infrared (NIR) camera. Moreover, machine learning enables reagent identification in real-time and over a wide dynamic range. Furthermore, by plotting the identification results as pixel values, the spatial distribution of reagents can be imaged at high speed without the need for post-processing.