Comparison of variable selection algorithms on vis-NIR hyperspectral imaging spectra for quantitative monitoring and visualization of bacterial foodborne pathogens in fresh pork muscles

Abstract

This research aims to verify the feasibility of developing an improved and efficient reduced spectrum model for quantitative tracking of foodborne pathogens. Rapid monitoring of bacteria foodborne pathogen (Escherichia coli O157 and Staphylococcus aureus) contamination of fresh longissimus pork muscles was implemented by employing visible near-infrared (Vis-NIR) hyperspectral imaging spectra and partial least squares regression algorithm (PLSR). Six (6) wavelength variables selection algorithms were applied to the full spectral information to determine the wavelength variables of the collected HSI spectra that provides essential and relevant information about the concentration of bacterial foodborne pathogen. Commonly used algorithms based on model population analysis (MPA) (2), Intelligent Optimization Algorithms (2), and Hybrid variable selection methods (HVSM) (2) were utilised to select characteristic wavelengths. Compared to other strategies, variable combination population analysis with genetic algorithm (VCPA – GA), and variable combination population analysis with iteratively retaining informative variables (VCPA – IRIV) considerably bettered the predictive efficiency of the model, suggesting that the updated VCPA step is a very efficient way to remove unrelated variables. Vcpa-based hybrid strategy is an effective and reliable approach for variable selection of visible near-infrared (vis-NIR) spectra. Visualising bacterial foodborne pathogen distribution map on the pork samples provided a more insightful and detailed evaluation of the bacterial contamination at each pixel, offering a novel approach for evaluating bacterial contamination of agricultural products.