Speaker
Description
The growing availability of portable, field-deployable spectrometers has expanded the range of applications in spectral analysis, but it also presents a significant challenge: measurements of the same sample obtained from instruments with different spectral windows and resolutions are often not directly comparable, limiting the ability to build robust and generalizable classification models. These issues are particularly critical in forensic analysis, such as the detection of psychotropic substances, which is the focus of the European research project NARCOSIS. In this framework, using the available literature data, we present a machine learning-based classification tool, aimed at supporting forensic operators in the identification of psychotropic substances from IR absorption spectra acquired. To address inter-instrument variability arising from differences in spectral range and resolution among heterogeneous spectrometers, spectra are pre-processed through linear interpolation onto a common wavelength grid prior to classification. This alignment step enables a single fully connected neural network classifier to be trained and evaluated on spectra originating from different source instruments, without requiring instrument-specific models. The proposed approach represents a scalable and practical solution for multi-instrument IR spectral analysis in forensic contexts, with promising potential for deployment within the NARCOSIS platform for real-world drug detection scenarios.
Project NARCOSIS has received funding from the European Union’s Horizon Europe research and innovation programme (Civil Security for Society) under grant agreement No 101168195.