Applications of Spectral Imaging to Monitor Poultry Product Quality and Detect Foodborne Pathogens Using Multuvariate Statistical Methods

To ensure consumer satisfaction and safety, poultry products must be routinely checked for authenticity, quality, microbiological spoilage, and contamination. Novel imaging techniques, such as spectral imaging, are being developed and employed to acquire real-time chemical and spatial information about products without destruction of samples to ensure safety of products and prevent economic losses. Further, recent advances in spectral imaging have allowed for the development of miniaturized portable instruments, which are more convenient for use in a food processing environment and less expensive than conventional instruments. This thesis explores applications of spectral imaging to poultry quality and foodborne pathogen detection as an alternative monitoring method. The literature review in Chapter 2 identified freeze-thaw identification, shelf-life prediction, and bacterial pathogen detection at the macroscopic level as applications requiring further research. To address these gaps, the experimental chapters of this thesis investigated the potential of spectral imaging to assess poultry product quality and identify foodborne bacterial contamination. Specific applications explored include: determining freeze-thaw authenticity, detecting products stored beyond their recommended “use-by” date, predicting microbiological spoilage, examining effects of UV decontamination, and identifying foodborne bacteria species dried on stainless steel. In Chapter 3, freeze-thaw authentication was more successful using VNIR than SWIR spectral imaging. Chicken thighs with skin on were the most suitable samples for detecting previously frozen products. Detection in the VNIR range was possible due to subtle changes in sample colour which occurred due to increased oxymyoglobin concentration following freeze-thaw cycles and were quantifiable by analysing VNIR spectra of the samples. Poor discrimination in the SWIR region was likely a result of increased absorbance by surface moisture due to drip loss from samples throughout the duration of the experiment. Research on freeze-thaw authentication was furthered in Chapter 4, using low cost and rapid portable spectral imaging in the visible spectral range and LED lighting. Freeze-thaw history of chicken thighs with skin on could be reliably detected through MAP plastic packaging. Detection with high accuracy was possible using full datasets and small randomly selected subsets of data, implying that subsampling could be sufficient for large scale industrial applications to decrease computational costs. Portable spectral imaging and LED lighting was also successfully used to predict the “use-by” status of chicken thighs with skin on through MAP plastic packaging with higher accuracy than RGB imaging in Chapter 5. Wavelengths associated with myoglobin derivatives (445 – 635 nm) were identified as pertinent to spoilage determination. In Chapter 6, it was found that UV decontamination increased a* values of chicken breast samples in vacuum packaging and further analysis of visible spectra, taken together with results from previous chapters and the literature, suggested that UV treatment accelerated oxymyoglobin formation. Despite differences in oxymyoglobin concentrations, it was possible to predict Total Viable Count (TVC) and Total Enterobacteriaceae Counts (TEC) of all samples (i.e. models included UV treated and untreated samples as one group). Finally, portable VNIR imaging was used in Chapter 7 to successfully discriminate between Gram type and within gram positive bacteria species using PLS-DA and SVM classifiers. For gram negative species, SVM was more successful at discriminating between species. Wavelengths associated with successful discrimination were mainly in the visible light region, opening to door to compatibility with low cost LED lighting.