The objective of this thesis was to investigate the feasibility and potential of using specific biosensor-based methods - electronic nose (e-nose) and Raman spectroscopy - as alternatives to conventional methods for food safety analysis. The principal shortcoming associated with conventional methods is the length of time required for an unequivocal result (5-7 days). Measurement systems were designed by taking raw sensor responses from off-the-shelf biosensor instruments, extracting meaningful features from these (using appropriately selected signal processing algorithms), and performing pattern classification (and validation) to assess their effectiveness in identifying the contaminant.
For bacteria (E. coli and Listeria) hosted in nutrient broth, e-nose based systems using metal oxide sensors (MOS) and fingerprint mass spectrometry were used to discriminate between bacteria at different concentrations with discrimination evident at levels down to 106 cells/mL.The capacity of these systems to identify bacteria at lower concentrations was hindered by a time-varying drift inherent to the nutrient broth.
Using individual bacterial colonies, a MOS-based e-nose measurement system was able to discriminate between four bacteria types (two E. coli strains, Listeria and Enterococcus). Multiple consecutive e-nose responses were employed to achieve classification accuracies exceeding 90%. A surface-enhanced Raman spectroscopy based measurement system employing wavelet-based pre-processing and feature extraction stages was developed to discriminate between species of Listeria, yielding classification accuracies of over 90%.
Motivated by the potential to incorporate e-nose technology in a smart home environment, a MOS-based system was used to measure food spoilage. The e-nose sensor responses demonstrated smooth trajectories in both original and dimension-reduced feature spaces and the degree of spoilage (as measured by sample age) were found to correlate with extracted features. In order to guide future research in this area, a framework for incorporating odour sensing into a smart home for identification of problems and activities of daily living is proposed.
Because the sample types used in this thesis are available relatively early in the inspection process, the performance of these methods justifies further work to assess their potential in a pre-screening capacity. The use of confidence measures is also explored to identify which samples might be classified with confidence with the proposed methods.