Abstract: The traditional method for detecting the total viable count of bacteria in liquid egg is laborious and time consuming. To overcome these drawbacks, the present study was undertaken to develop a fast method for predicting the total viable count of bacteria in liquid egg by using hyperspectral image technology (400–1 000 nm). The hyperspectral images of artificially inoculated liquid egg samples with different contamination levels of Pseudomonas aeruginosa were acquired. Then successive projections algorithm (SPA) was used to extract feature wavelengths, and partial least squares (PLS) and support vector machine (SVM) models were developed based on the feature wavelengths and the full spectra, respectively. Finally, the performance of the multivariate prediction models were compared and analyzed. The result showed that the Autoscale method was the best pretreatment method and the SVM model was the best prediction model for the total viable count of bacteria in liquid egg. The correlation coefficient of prediction (RP) and the root mean square error of prediction (RMSEP) was 0.81 and 0.63 (lg(CFU/g)) for egg white, 0.82 and 0.47 (lg(CFU/g)) for egg yolk, 0.75 and 0.75 (lg(CFU/g)) for liquid whole egg, respectively. Overall, hyperspectral imaging combined with chemometrics enables quantitative prediction of the degree of microbial contamination in liquid egg.

Key words: liquid egg, total viable count of bacteria, hyperspectral imaging technology, model, fast prediction