Evaluating near infrared hyperspectral imaging as a complimentary rapid screening tool for food microbiology

Simple item page
dc.contributor.advisorWilliams, Paul Jamesen_ZA
dc.contributor.authorMapling, Celeste Nadineen_ZA
dc.contributor.otherStellenbosch University. Faculty of AgriSciences. Dept. of Food Science.en_ZA
dc.date.accessioned2020-02-14T13:38:06Z
dc.date.accessioned2020-04-28T15:10:27Z
dc.date.available2020-02-14T13:38:06Z
dc.date.available2020-04-28T15:10:27Z
dc.date.issued2020-03
dc.descriptionThesis (MScAgric)--Stellenbosch University, 2021.en_ZA
dc.description.abstractENGLISH ABSTRACT: Near infrared hyperspectral imaging (NIR-HSI) was evaluated as a complimentary, rapid screening tool for microbiology. Six bacterial isolates were used including Bacillus cereus (ATCC 13061), Escherichia coli (ATCC 25922), Salmonella enteritidis (ATCC 13076), Staphylococcus aureus (ATCC 29213 & 25923) and Staphylococcus epidermidis (ATCC 12228). The bacteria were streaked out onto nutrient agar (NA) and tryptic soy agar (TSA) to evaluate the effect of different growth media on the NIR spectra. For the second objective both a streak and spread plate method was used on only NA to evaluate the effect of the plating methods on the spectra. The bacteria were classified based on their Gram-stain classification (Gram-positive or Gram-negative) and based on pathogenicity (pathogenic and non-pathogenic). Hyperspectral image analysis was conducted in the 950-2500 nm range. The images were collected using the Hyspex SWIR-384 push-broom imaging system. Two pre-processing methods were applied including standard normal variate (SNV) and Savitzky-Golay (2nd derivative, 3rd polynomial,) to assist with image cleaning. The images were mosaicked into groups based on the overall objective of the analysis. Principal component analysis (PCA) models were calculated for initial data exploration and reduction of data dimensionality. Partial least squares discriminant analysis (PLS-DA) models were then calculated to classify the individual bacteria based on Gram-stain reaction and pathogenicity. The PLS-DA model of the bacteria streaked out on TSA produced the best overall classification accuracy (100%). The bacteria on NA obtained a classification accuracy of 99.55% and the Gram-stain classification and pathogenicity models produced classification accuracies of 98.29% and 98.72%, respectively. For the second objective the bacteria were classified based on Gram-stain reaction. To further optimise the model, three different pre-processing combinations were employed. The PLS-DA model containing only Gram-negative bacteria achieved a 100% classification accuracy, followed by 99% for Gram-positive bacteria and 89% for the model combined. This proves that hyperspectral imaging can successfully be implemented as a rapid screening tool to accurately detect and identify foodborne pathogenic bacteria.en_ZA
dc.description.abstractAFRIKAANSE OPSOMMING: Geen opsomming beskikbaar.af_ZA
dc.description.versionMastersen_ZA
dc.embargo.terms2020-12-31
dc.format.extentxv, 108 pages : illustrationsen_ZA
dc.identifier.urihttp://hdl.handle.net/10019.1/108344
dc.language.isoen_ZAen_ZA
dc.publisherStellenbosch : Stellenbosch Universityen_ZA
dc.rights.holderStellenbosch Universityen_ZA
dc.subjectHyperspectral imagingen_ZA
dc.subjectNear infrared spectroscopyen_ZA
dc.subjectFood -- Microbiologyen_ZA
dc.subjectBacterial growthen_ZA
dc.subjectFoodborne diseasesen_ZA
dc.subjectUCTD
dc.titleEvaluating near infrared hyperspectral imaging as a complimentary rapid screening tool for food microbiologyen_ZA
dc.typeThesisen_ZA
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
mapling_infrared_2020.pdf
Size:
6.88 MB
Format:
Adobe Portable Document Format
Description:
Download
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
1.71 KB
Format:
Plain Text
Description:
Download
Collections
Masters Degrees (Food Science)