Machine learning combined with multi-source data fusion for rapid quality assessment of yellow rice wine with different aging years

Abstract

Aging is a crucial factor for high-quality alcoholic beverages, which improves the flavor of alcoholic beverages such as yellow rice wine (YRW). However, traditional analytical methods for evaluating the quality of YRW are very tedious, and there is still a lack of rapid quality assessment methods for YRW with different aging times. In this study, taking Jimo rice wine (JRW, a representative YRW in northern China) with different aging years as the object, traditional analytical methods and various antioxidant experiments were used to characterize their quality differences, a rapid quality evaluation method was established by machine learning combining with the data obtained from Flash GC e-nose and near-infrared. The results showed multiple physicochemical parameters coupled with multivariate statistical analysis could distinguish JRW with diverse aging years. JRW with different aging years also showed various antioxidant capacities and generally increased with the aged years. A total of 24 major aroma components were identified in JRW, of which five varied regularly with aging time. Then, the deep learning algorithm (long short-term memory, LSTM) showed excellent classification performance (100% accuracy) in JRW with different aging years, and a multi-source information fusion strategy can achieve 100% classification accuracy even when combined with traditional algorithms. Finally, the fused data improved the accuracy of the LSTM regression model in predicting the content of the main physicochemical parameters of JRW, with higher R2 and lower RMSE compared to data from a single source. Overall, this study clarified the quality differences of JRW with diverse aging years, and a rapid and precise method combining multi-source data fusion and machine learning was developed to assess the quality of JRW, which could also apply to other beverages or foods.