Non-destructive detection of TVC in pork by machine learning techniques based on spectral information

Abstract

The rapid and nondestructive identification of pork spoilage holds significant importance due to the inherent richness of nutrients and the conducive environment for bacterial proliferation within pork. This study focused on the non-destructive assessment of the total viable count in pork utilizing visible/near-infrared spectroscopy. By employing this technique, pork samples were subjected to analysis across the visible/near-infrared spectra range (400-1000 nm), with the total viable count determined through the plate counting method. The principal component analysis technique was used to consider whether there was variability in the spectra of pork with different total viable counts. Three different preprocessing methods in visible near-infrared spectroscopy for the prediction of total viable count in pork were compared, the preprocessing methods used were standard normal variate, multiplicative scatter correction and Savitzky-Golay smoothing. The results of the study show that, divisibility of pork with different total viable count in the low-dimensional space of the first and second principal components of principal component analysis. Among these preprocessing techniques, the study highlighted the superiority of the partial least squares regression model combined with standard normal variate preprocessing. This optimized model exhibited remarkable efficiency in predicting total viable count in pork. The best total viable count prediction model showed the RP, RMSEP, and RPD of 0.864, 0.826%, and 1.887, respectively. This study highlights the importance of rapid and non-destructive techniques for pork spoilage detection, contributing to improved food safety and quality assurance practices within the pork industry.