Artificial intelligence application built for ATS detection with a new portable hollow fiber IRAS spectrometer

Abstract

We are presenting signal preprocessing and analysis methods that were successfully applied for improving the performance of a new portable hollow fiber infrared absorption spectrometer built for screening for amphetamine-type stimulants (ATS) in law enforcement operations. The challenge in developing the automated class identity assignment system was to model the similarity of the spectra of amphetamines and, in the same time, to discriminate them from non-amphetamines based only on the absorptions recorded in a very narrow spectral window (1550 -1330 cm -1 ), as the radiation source of the portable spectrometer is a Quantum Cascade Laser. The presented signal selective amplifier has significantly enhanced the efficiency of the developed artificial intelligence application screening for the targeted drugs of abuse. The main variables contributing to the formation of distinct clusters (associated to stimulant and to hallucinogenic amphetamines) has been performed by Principal Component Analysis. The classification efficiency has been evaluated based on Hierarchical Cluster Analysis, i.e. agglomerative clustering.