Non-destructive classification of dehydrated quince slices using image processing and comparative machine learning approaches

Seda Günaydın,
Non-destructive classification of dehydrated quince slices using image processing and comparative machine learning approaches,
Journal of Food Composition and Analysis,
Volume 148, Part 3,
2025,
108469,
ISSN 0889-1575,
https://doi.org/10.1016/j.jfca.2025.108469.
(https://www.sciencedirect.com/science/article/pii/S0889157525012852)
Abstract: Advanced analytical techniques improve the accuracy and efficiency of classifying dehydrated fruits for food processing and packaging. This study categorized dried quince samples based on their chromatic properties under twelve different drying conditions and pretreatment combinations. Three different microwave power levels (300, 600, and 900 W) were applied in the drying process, and the fruit slices were pretreated with ultrasound, 5 % ascorbic acid solution, mandarin juice, or no pretreatment. The color channels (R, G, B, L, a, b, H, S, and V) were extracted from images of the dehydrated quince slices. This study evaluated the performance of classification models that integrate color channels with artificial intelligence methods. The RF, BAG, and PART models achieved overall accuracies of 99.67 %, 99.17 %, and 99.00 %, respectively, across all color channels. Dried quince slices exhibited the highest L value in the 300 W-MJ (79.71) treatment group. The 600 W-MJ group had the highest a value (10.16), while the 600 W-AA group exhibited the maximum b value (60.14). The highest R and V values were observed in the 900 W-C (230.66) and 300 W-MJ (231.12) groups, respectively.
Keywords: Dehydrated; Quince; Image processing; Machine learning; Classification