Food traceability analysis and quality marker monitoring in Ophiocordyceps sinensis with multimodal ensemble learning.

Ophiocordyceps sinensis (OS) faces serious risks of food fraud, including quality misrepresentation, adulteration and illegal additives. To preserve the economic interests of consumers and the transparent management of food trade, so this study proposed a rapid and non-destructive detection tool to identify traceability of the growth environment and predict quality markers of OS. Colors, textures and spectra were utilized to build unimodal models, respectively. In addition, Bayesian optimization was utilized to automatically find the optimal hyperparameters. A multimodal ensemble learning framework was developed and evaluated against unimodal model. The proposed system demonstrated improvements. A 10.34 % enhancement in the accuracy was achieved for traceability. Also, the RPD of the predictive models for polysaccharide and nucleoside were increased by 9.88 % and 3.49 %, respectively. The system as a potential "spectral eye" technology for non-destructive prediction of OS food traceability and quality markers monitoring.