A computational study of spectral matching algorithms for identifying Raman spectra of polycyclic aromatic hydrocarbons

Abstract

To facilitate experimental development of a Raman‐based chemical sensor for identifying 16 carcinogenic polycyclic aromatic hydrocarbons (PAHs) in food and selection of an appropriate spectral matching algorithm for use with the sensor, computer simulations are carried out for evaluating the performance of several spectral matching algorithms in identifying Raman spectra of target PAHs in the presence of strong interference from co‐existent PAH spectra. The studied algorithms are the following: the Pearson correlation coefficient, the Euclidean distance, and the cosine distance in the spectral space and in a normalized principal component space (CD‐NPCA). The simulations are performed with mixture Raman spectra synthetized from a reference Raman spectral library of the 16 PAHs in the 1000–1700 cm−1 fingerprint spectral range that is calculated using density functional calculations. Receiver operating curves are generated for each target PAH and spectral algorithm pair to assess the performance of the algorithms. It is shown from the study that the CD‐NPCA outperforms the others in terms of speed and discriminating power for identifying the target spectra in the mixture spectra due to dimensionality reduction and an angular augmentation effect of input spectral data. This study provides a cost‐effective way for designing the Raman‐based sensor for PAH detection and paves the way for future experimental development of such a sensor. Copyright © 2016 John Wiley & Sons, Ltd.