AI-driven multimodal optical sensing for identification of foodborne bacterial species

Traditional methods for bacterial detection and identification are labor-intensive, time-consuming, and often require specialized laboratory environments, limiting their suitability for on-site food safety monitoring. This study developed a method to identify common foodborne bacterial species using a recently developed compact multimodal optical sensing system with integrated fluorescence, color, and transmission imaging, and Raman spectroscopy for rapid, automated, and intelligent food safety inspection. Five bacterial species were selected for testing the system: Bacillus cereus , Escherichia coli , Listeria monocytogenes, Staphylococcus aureus , and Salmonella enterica . A total of 600 nonselective agar plates (120 plates/species), representing eight strains per species, were tested. Using the system’s automated sampling function, multimodal images were acquired from each plate and Raman spectra were collected from each colony. Artificial intelligence models were developed to differentiate the five bacterial species based on the multimodal image and spectral datasets. On a test set comprising 60 agar plates, the You Only Look Once version 11 n (YOLOv11n) models achieved classification accuracies of 98.2%, 98.8%, and 100% for the fluorescence, color, and transmission images, respectively, while a support vector machine classifier achieved 99.0% accuracy using the Raman spectra. A decision-level fusion of the four models achieved 99.6% test-set classification accuracy. The AI-driven multimodal optical sensing approach demonstrates strong potential for rapid and accurate identification of bacterial species following incubation, offering a promising tool for automated microbial screening in food safety inspection settings.