Automatic recognition of food bacteria using Raman spectroscopy and chemometrics: A comparative study of multivariate models

Abstract

Food safety is the foundation of trust for food stakeholders. Contamination, especially from biological sources, at food processing plants can threaten this foundation, resulting in negative impacts on consumer health and substantial economic losses. Therefore, a rapid, effective, and noninvasive method for detecting bacteria in the food industry is essential. In this study, Raman micro-spectroscopy with advanced statistical tools is proposed as a mean of detecting and differentiating between various types of bacteria. This approach circumvents the complexities of traditional culture-based detection methods. Specifically, fifty-two bacterial strains of 39 different genera were analyzed using Raman spectroscopy. As a result, about 2563 Raman spectra were generated and integrated into the database. This huge amount of spectral data was analyzed using several chemometric tools, including principal component analysis (PCA), factorial discriminant analysis (FDA) k-nearest neighbors’ algorithm (KNN), and convolutional neural network (CNN). Our multivariate data analysis showed that the developed method is rapid and capable of distinguishing several strains. While, FDA models showed mediocre performances, KNN models provided good bacterial classification for most of the analyzed strains (average correct classification 90–95%). In comparison, CNN achieved a higher classification accuracy, of 97%, compared with other models. Combining Raman spectroscopy with chemometric tools yields a robust bacterial assessment method that is simple, rapid, and efficient.