Value-Added Production and Process of the Us Midwestern Grapes and Development of Machine-Learning-Driven Raman Spectroscopy for Rapid Detection of Edible Oil Quality

This dissertation studied three independent but related objectives. The first objective focused on the value-added production of berries of five major Midwest grape cultivars by enriching the micronutrients, Selenium and Lithium in the fruits to improve their nutraceutical properties. Results demonstrated agronomic biofortification increased the Selenium and Lithium content in the whole grape by multiple times; meanwhile, it limitedly affected the vine vigor and fruit quality. This study demonstrated agronomic biofortification was an effective strategy to biofortify micronutrients in grape berries. Therefore, it could be a promising strategy to increase the nutrition and consumption of Midwest grapes. Grape pomace is a winery waste but rich in polyphenols, which were reported as natural antioxidants to inhibit lipid oxidation in foods, therefore increased the value of pomace. However, contradictory conclusions were also usually made by different studies. For an effective value-added process of grape pomace as natural antioxidants, the second objective of this dissertation was to understand the mechanism of polyphenols in preventing lipid oxidation under different models (chemical solution, emulsions, and mayonnaise) and conditions (pH, iron, protein interaction). Key findings indicated, at neutral pH, chelation was the crucial ability that enabled polyphenols to inhibit lipid oxidation (\u3e90%). But at acidic pH, other activities (e.g., hydrogen donation, reduction) rather than chelation allowed polyphenols to partially inhibit lipid oxidation (40%-60%). This study advanced the understanding of polyphenols in preventing lipid oxidation. When evaluating lipid quality, we realized the traditional methods were time-consuming. Therefore, the third objective was to develop a rapid, accurate, and cost-effective way for lipid quality analysis by coupling machine-learning with Raman spectroscopy. Results showed machine-learning-driven Raman spectroscopy obtained an accuracy of 96.7% in detecting oil type and an adulteration prediction rate of R2 ≤ 0.984. The oxidation prediction rate was R2 ≤ 0.8180, which is subject to improvement. Overall, this study provided an exemplar for applying machine learning and Raman spectroscopy in the rapid detection of lipid quality