Journal of Spectral Imaging,   Volume 8   Article ID a5   (2019)

Peer reviewed Paper

Effect of colony age on near infrared hyperspectral images of foodborne bacteria

  • Paul J. Williams
  • Terri-Lee Kammies
  • Pieter A. Gouws
  • Marena Manley
Department of Food Science, Stellenbosch University, Private Bag X1, Matieland (Stellenbosch) 7602, South Africa

 https://orcid.org/0000-0001-5382-8395
 Search for papers by this author
Department of Food Science, Stellenbosch University, Private Bag X1, Matieland (Stellenbosch) 7602, South Africa

 https://orcid.org/0000-0002-7061-2403
 Search for papers by this author
Department of Food Science, Stellenbosch University, Private Bag X1, Matieland (Stellenbosch) 7602, South Africa

 https://orcid.org/0000-0001-7581-7208
 Search for papers by this author
 Corresponding Author
Department of Food Science, Stellenbosch University, Private Bag X1, Matieland (Stellenbosch) 7602, South Africa
[email protected]
 https://orcid.org/0000-0002-6014-2049
 Search for papers by this author

Near infrared hyperspectral imaging (NIR-HSI) and multivariate image analysis were used to distinguish between foodborne pathogenic bacteria, Bacillus cereus, Escherichia coli, Salmonella Enteritidis, Staphylococcus aureus and a non-pathogenic bacterium, Staphylococcus epidermidis. Hyperspectral images of bacteria, streaked out on Luria—Bertani agar, were acquired after 20 h, 40 h and 60 h growth at 37 °C using a SisuCHEMA hyperspectral pushbroom imaging system with a spectral range of 920–2514 nm. Three different pre-processing methods: standard normal variate (SNV), Savitzky—Golay (1stderivative, 2nd order polynomial, 15-point smoothing) and Savitzky—Golay (2nd derivative, 3rd order polynomial, 15-point smoothing) were evaluated. SNV provided the most distinct clustering in the principal component score plots and was thus used as the sole pre-processing method. Partial least squares discriminant analysis (PLS-DA) models were developed for each growth period and was tested on a second set of plates, to determine the effect the age of the colony has on classification accuracies. The highest overall prediction accuracies where test plates required the least amount of growth time, was found with models built after 60 h growth and tested on plates after 20 h growth. Predictions for bacteria differentiation within these models ranged from 83.1 % to 98.8 % correctly predicted pixels.

Keywords: colony age, foodborne bacteria, growth media, PLS-DA, NIR hyperspectral imaging

Metrics

Downloads:

149

Full-Text HTML Views:

28

Abstract Views:

785