Selection of Informative Spectral Wavelength for Evaluating and Visualising Enterobacteriaceae Contamination of Salmon Flesh

摘要

Enterobacteriaceae is one kind of harmful microorganisms commonly presented in raw fish products, and detection of Enterobacteriaceae plays a very important role in evaluating microbial contamination. This work was carried out to exploit the potential of emerging hyperspectral imaging technique to determine the Enterobacteriaceae contamination of salmon flesh during cold storage. The spectral information ranging from 900 to 1700 nm (239 wavelengths) was extracted to relate to the Enterobacteriaceae loads (recorded as log 10 CFU/g) using partial least square (PLS) regression, developing a PLS model with correlation coefficient of prediction (r (P)) of 0.94 and root mean square error of prediction (RMSEP) of 0.53 as well as residual predictive deviation (RPD) of 2.97. By applying successive projection algorithm (SPA), eight wavelengths at 924, 931, 964, 1068, 1262, 1373, 1628 and 1668 nm among the 239 wavelengths were selected as informative wavelengths to reduce the information redundancy and optimise the PLS model. With the eight informative wavelengths, a simplified PLS model defined as SPA-PLS was established with r (P) of 0.95, RMSEP of 0.47 and RPD of 3.23. To visualise the contamination degree of salmon flesh caused by Enterobacteriaceae, the SPA-PLS model was transferred to each pixel of images, and colourful distribution maps were produced with different colour represented different numbers of Enterobacteriaceae colonies. The results showed that hyperspectral imaging operating in 900-1700 nm is promising in evaluating Enterobacteriaceae contamination of salmon products. More studies are still required to further refine the multispectral imaging system to achieve online application.