Automated identification of illicit substances by chemiometrics. Application to FTIR and RAMAN analysis.
Abstract
The CEA-List is currently developing, in collaboration with the Laboratoire Central de la Préfecture de Police de Paris (LCPP), a decision-making software for the analysis of hazardous substances using data from FTIR, RAMAN and X-ray fluorescence spectrometers. These equipments are well suited to provide rapid analysis of unknown organic or mineral substances encountered in the field. One of the limitations of this type of analysis is the interpretation of the results when compounds are present in a mixture. Indeed, when an unknown product contains several substances, the spectrum obtained is a combination of the spectra of each of these substances, which complicates the experts interpretation. The identification may even be impossible if the target substance is present in very small amounts in the mixture.
The proposed tool is based on the use of supervised learning algorithms, designed to solve data classification problems [1]. It recognizes the target substances from a threat database provided by the user. It automatically generates mixture spectra from pure compound spectra that are used in the training step. Then, it is able to identify or rule out the presence of a threat and also to provide a confidence score in the response provided. Finally, it offers an easy-to-use graphical interface.
Currently, we are looking into solutions to automatize the classifier parametrization, in order to limit the number of inputs required from users and make it more user-friendly.