Non-destructive biochemical profiling of goldenberry ripening using confocal micro Raman spectroscopy coupled with machine learning.

Abstract

The accurate monitoring of the fruit ripening is crucial for maintaining quality and prevention from post-harvest economic losses. Goldenberry, being rich in vitamins, minerals, and carotenoids, yet its ripening assessment relies on variable and often subjective traditional approaches. To address this, the present study utilises rapid and non-destructive techniques using confocal micro-Raman and UV-Vis spectroscopy to track biochemical changes in both the exocarp and mesocarp of goldenberry fruit across five ripening stages: pre-mature, mature, pre-ripe, ripe, and post-ripe. The analysis of Raman spectra of exocarp shows an increase in carotenoid bands up to ripe stage, followed by a decrease at post-ripe stage, while the mesocarp exhibits a continuous increase in carotenoid intensity through stage post-ripe. The UV-Vis spectral analysis supports these observations, indicating increasing carotenoid content alongside a reduction in chlorophyll as ripening progresses. Among different machine learning (ML) approaches, unsupervised clustering demonstrates strong performance of k-means and agglomerative nesting, with high rand index values (>0.97). In supervised domain, support vector machine, logistic regression, random forest, decision tree, and principal component-linear discriminant analysis achieves the highest values of performance metrices including accuracy, precision, recall, F1 score, ROC/AUC values, and cross validation scores. Moreover, partial least squares regression analysis demonstrates excellent predictive power, with R² = 0.93 (calibration) and 0.70 (cross-validation) for exocarp and R² = 0.91 (calibration) and 0.89 (cross-validation) for mesocarp. The combined use of Raman with ML enables a practical and reliable framework for assessing fruit ripening, with promising applications in quality control and post-harvest management.