Chemical Identification from Raman Peak Classification Using Fuzzy Logic and Monte Carlo Simulation

Abstract

In spite of the wide use of Raman spectroscopy for chemical analysis in different fields, not any automated identification of Raman spectra is universally adopted. However, the interest in this field is witnessed by the large number of papers published in the last decades. The problem of Raman-spectra classification becomes particularly challenging when low irradiation is requested, either for safety reasons or to avoid target photodegradation. This often leads to spectra characterized by a low signal-to-noise ratio, where methods based on correlation usually fail. For this reason, a method based on peak identification through FMFs is presented, discussed and validated over a large set of samples. In particular, a Monte Carlo simulation has been employed to determine the best parameters of the fuzzy membership functions based on the analysis of performances of the classification procedure. The ROC curves have been analyzed, and AUC and best accuracy are employed as key parameters to evaluate the classification performances on different amounts of ammonium nitrate (from 300 to 1500 μg) and different laser exposure levels (from 3.1 to 250 mJ/cm2).