You are here : Home > Automating toxic chemical identification


Smart digital systems

Automating toxic chemical identification

List, a CEA Tech institute, partnered with the French National Center for Scientific Research (CNRS) to develop an algorithm to automatically identify toxic chemicals and explosives in real time. The research was part of a French government CBRN-E 1 counterterrorism program.

Published on 6 June 2019

The Police Nationale, one of France’s national police forces, turned to List, a CEA Tech institute, to develop a software application to automatically identify toxic chemicals and explosives in real time. The software leverages a support vector machine (SVM), a sophisticated tool used in artificial intelligence techniques just like neural networks.  

Unlike neural networks, however, SVMs are optimal classifiers. In other words, they can determine with a high degree of confidence whether or not a chemical compound is present in a mix.

Any toxic compounds present have to be identified one after the other. So, the researchers developed a smart technique to preselect areas of interest in the FT-IR spectra of compounds included in a database of threats. These areas of interest are then inspected within the spectrum of an unknown sample. The information provided to the SVM is pre-calculated and simplified using conventional signal analysis methods. 

Another original aspect of the software is that the SVM learns from theoretical spectra, eliminating the need for it to learn from huge databases. A tool developed by List can automatically generate, from just a few real spectra, theoretical spectra of chemical mixes from the thousands of molecules.

Blind tests revealed excellent selectivity and sensitivity (to within 4% or 5% for chemical mixtures in powder form). These methods, called Peak Correlation Classification (PCC), were patented by List. 

[1] Chemical, biological, radioactive, nuclear, and explosives
[2] FT-IR, or Fourier Transform Infrared Spectroscopy 

Top page

Top page