Abstract: This study aimed to simultaneously and rapidly identify and count the common fermentation strains (Lactobacillus bulgaricus and Streptococcus thermophiles) and probiotic strains (Lactobacillu caseii, Lactobacillus acidophilus and Lactobacillus plantarum) in probiotic yogurt by hyperspectral imaging technology combined with plate colony method using pattern recognition. The obtained results were compared with those obtained by the traditional counting method to evaluate the feasibility of colony counting using hyperspectral imaging technology. It was shown that when nine principal components were used, the least square support vector machine (LS-SVM) model was superior to the K-nearest neighbor (KNN) and back-propagation artificial neural network (BP-ANN) models in identifying the different types of colonies. The recognition rate of the LS-SVM model was 99.20% and 93.33% for calibration and prediction sets, respectively. Thus, the LS-SVM model was the best counting model. No significant difference was observed between the two counting methods (P > 0.05), confirming the feasibility of using hyperspectral imaging technology to simultaneously count the number of each probiotic species in yogurt with mixed strains. The method overcame the defect of the traditional counting method which could not simultaneously count the number of each probiotic species in yogurt mixed strains, and therefore could provide the basis for rapid and non-destructive detection of the quality and health benefits of yoghurt.

Key words: hyperspectral imaging technology, yogurt, pattern recognition, colony counting, probiotic