A rapid, non‐destructive, and accurate method for identifying citrus granulation using Raman spectroscopy and machine learning

Abstract Citrus fruits are widely consumed for their nutritional value and taste; however, juice sac granulation during fruit storage poses a significant challenge to the citrus industry. This study used Raman spectroscopy coupled with machine learning algorithms to rapidly, non‐destructively, and precisely detect citrus granulation. The investigation analyzed 969 Raman spectral data points, comprising 714 non‐granulated and 255 granulated citrus samples. Following logistic regression, decision tree, and partial least squares discriminant analyses, the optimal model was refined using principal component analysis, a successive projection algorithm, and a competitive adaptive reweighted sampling algorithm (CARS). The identified characteristic Raman peaks at certain wavenumbers were used as input data for the classification model, revealing differences in the water, ferulic acid, and sugar contents between granulated and non‐granulated samples. The partial least squares discriminant classification model achieved an accuracy rate of 0.997, recall rate of 0.994, and F ‐fraction of 0.996 after preprocessing the standard deviation data and selecting 22 optimal principal components. The critical peaks extracted from the citrus Raman spectra were those at wavenumbers of 1580 and 1661 cm −1 . The classification model based on combined second derivative—CARS—partial least squares discriminant analysis exhibited the best performance, achieving 100% accuracy for all test sets. The proposed method provides a scientifically robust and reliable means of assessing the quality of an entire citrus crop. Reduced wastage and economic losses, and the related environmental effects of food waste. Practical Application The proposed methods can determine if citrus fruit has become granulated during storage. Additionally, they provide technical support for screening granulated citrus in a pipeline, thereby providing a more scientific and reliable classification of the quality of a citrus crop.