Document Type : Research Article-en
Authors
1
Research Scholar, Department of Mechanical Engineering, Zeal College of Engineering and Research, Narhe Pune
2
Department of Mechanical Engineering, Maharshi Karve Stree Shikshan Sanstha's Cummins College of Engineering for Women, Pune-411052, Maharashtra, India
10.22067/ifstrj.2026.99699.1606
Abstract
The safety and identification of banana ripening treatments are important concerns in food supply chains, particularly because calcium carbide (CaC₂) may be used as an artificial ripening agent. This study examines the potential of non-destructive spectral fingerprinting for discriminating banana ripening treatments in Cavendish bananas (n = 40, 10 per treatment) observed over seven consecutive days, using both a four-class model (natural ripening, ethylene 15 g, calcium carbide 15 g, and calcium carbide 20 g) and a binary model (Natural vs. Artificially-Treated), with daily reflectance measurements collected using an AS7265X spectral sensor (410–940 nm). The spectral variance was distributed among the first two principal components (59.02%, PC1: 37.62%, PC2: 21.40%). The overall four-class accuracy of the Random Forest (RF) model was 77.14%, with its macro-average F1-score being 77.13%, when the banana level GroupKFold cross-validation was used to evaluate the model. The accuracy, F1-score and AUC for the three combined artificially treated groups were: Natural vs. Artificially-Treated = 91.79%, 94.59% and 0.958, respectively. In the four-class model, 485, 645, 560, 680, and 730 nm were the five highest-ranked wavelengths, whereas 560, 610, 645, 485, and 730 nm were the five highest-ranked wavelengths in the binary model. The results are preliminary evidence of the potential of compact multi-spectral sensing for non-destructive screening of banana ripening treatments.
Keywords
Subjects
