Rapid Quantification for the Incubation Quality of the Sweet Flour Sauce Koji by Machine Vision Combining Hyperspectral

ABSTRACT The koji quality was very important for the production of sweet flour sauce. In this study, an alternative method to chemistry analysis was established by combining machine vision and hyperspectral to quickly quantify the quality of sweet flour sauce koji (SFSK). The prediction models of the quality indicators were built by RF, SVM, and ELM with the raw data of machine vision and hyperspectral as well as the fusion data that was obtained by fusing the screened characteristic wavelengths with color variables, respectively. The results indicated that the accuracies of data fusion models were better than those models built with the separate data of hyperspectral and machine vision. Among all the models, the SVM model showed the best performance with the highest RPD value of 6.311 for total acid and the lowest RPD value of 3.037 for glucoamylase. This study offered the basis for real‐time monitoring of the SFSK production.