The rapid detection of food-borne pathogenic bacteria is critical to the food industry for preventing the introduction of contaminated product into the marketplace and limiting the spread of outbreaks. Hyperspectral microscope images (HMI) are a form of optical detection, which classify bacteria by combining microscope images with a spectrophotometer. The objective of this study was to compare the spectra generated from dark-field HMIs of five live Salmonella serotypes from two lighting sources, metal halide (MH) and tungsten halogen (TH), assessing classification accuracy and robustness, between 450 nm and 800 nm. It was found that the MH spectra could be reduced to as few as 10 optimal bands between 594 nm and 630 nm, but TH band reduction decreased accuracy, due to the inherent broader peak structure generated by the TH light source. Collection of HMIs from the two light sources comparing the same cells shows slight differences in scatter intensity patterns. Principal component linear discriminate analysis classified serotype subsets (n = 1800), reporting both MH and TH accuracies at 100%, while the reduced key MH bands achieved 99.4–100% accuracy. Principal component regression calculated the root mean squared error of cross-validation < 0.014 and a R2 > 0.948 for both full spectrum lamps. MH or TH lamps can be effectively used for discriminating bacteria HMIs on a cellular level by serotype, but reducing TH bands may lose crucial classification information.
Keywords: hyperspectral, microscopy, Salmonella, rapid detection, food safety, tungsten halogen, metal halide