A study on the off-flavor in the storage of citrus, based on Raman spectroscopy combined with machine learning

Abstract

Citrus fruits' off-flavor is a significant concern for consumers, highlighting the need for effective testing methods. Traditional detection techniques are often complex, time-consuming, and destructive. In contrast, Raman spectroscopy offers a rapid, precise, and non-destructive solution to these challenges. This study aims to investigate the effects of various storage conditions on the flavor quality of citrus fruits and to integrate machine learning models for rapid, non-destructive detection. Raman spectroscopic analysis revealed significant variations in the characteristic peaks of citrus essential oils (C-H, C-C, and CC bond vibrations) and sugars (C-H bending vibrations) at shifts of 756 cm⁻¹ , 1438 cm⁻¹ , 1602 cm⁻¹ , and 866 cm⁻¹ . In particular, off-flavor citrus exhibits significant changes in characteristic peaks, which related to changes in substance composition during flavor alteration. The intensity ratio of Raman characteristic peak indicates that D-limonene tends to degrade (I₁₄₃₈/I₁₅₂₉ decreases), while the content of α-terpineol tends to increase (I₁₆₀₆/I₁₅₂₉ increases) during the process of flavor quality change under different storage conditions. Machine learning results demonstrate that among the models used to identify off-flavor citrus, the Second-order Differentiation Support Vector Machine model performs optimally, achieving both accuracy and F-score of 100 %. This study provides technical support for optimizing citrus storage and promoting sustainable development within the industry.