Comparative Modelling of Walnut Kernel Weight Using Non-Destructive Physical and Mechanical Measurements
Tekirdağ Ziraat Fakültesi Dergisi · 22.05.2026 · 23 (3) · 832-849
DOI: 10.33462/jotaf.1642095 ↗Kurz gefasst
Fachartikel in der Zeitschrift Tekirdağ Ziraat Fakültesi Dergisi, online erschienen am 22.05.2026. Verfasst von Sefa Altıkat und 4 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben.
Abstract (Originalsprache)
A product undergoes several stages before reaching the consumer. Although cultivation is completed at harvest, ensuring that the product reaches consumers under optimal conditions is equally important. Postharvest processes, including transportation, storage, processing, and packaging, are influenced by numerous factors. Understanding the intrinsic characteristics of agricultural products is therefore essential for minimizing the adverse effects of these factors and for implementing appropriate management strategies. For many years, consumer preferences have primarily focused on the external appearance of fruits and nuts. In recent years, however, increasing attention has been directed toward internal quality attributes. Consequently, a major challenge for growers and marketers is the reliable assessment of internal quality characteristics without causing damage to the product. In this context, non-destructive techniques have gained considerable importance, as they enable the evaluation and prediction of internal quality parameters based on measurable external properties. The present study aimed to model walnut kernel weight using non-destructive approaches. To this end, several modelling techniques were evaluated, including multiple linear regression (MLR), principal component analysis (PCA), artificial neural networks (ANN), PCA combined with ANN (PCA+ANN), deep learning neural networks (DLNN), and support vector machines (SVM). The input variables included variety, walnut length (mm), width (mm), thickness (mm), walnut weight (g), geometric mean diameter (mm), sphericity (%), and cracking force (N). The results demonstrated that the SVM model achieved the highest predictive performance for walnut kernel weight estimation, with an accuracy of 98.14%. The ANN and DLNN models yielded accuracy values of 89.78% and 93.49%, respectively. In contrast, the hybrid PCA+ANN model exhibited a substantial decline in performance, achieving an accuracy of 58.8%. The MLR model produced an accuracy of 76.43%. Overall, the findings indicate that, among the evaluated machine learning techniques, the SVM approach provides the most accurate and robust predictions for non-destructive estimation of walnut kernel weight.
Zitieren
APA
Altıkat, S., Gülsoy, E., Kuş, E., Gulbe, A., & Altıkat, A. (2026). Comparative Modelling of Walnut Kernel Weight Using Non-Destructive Physical and Mechanical Measurements. Tekirdağ Ziraat Fakültesi Dergisi, 23(3), 832-849. https://doi.org/10.33462/jotaf.1642095
Vancouver
Altıkat S, Gülsoy E, Kuş E, Gulbe A, Altıkat A. Comparative Modelling of Walnut Kernel Weight Using Non-Destructive Physical and Mechanical Measurements. Tekirdağ Ziraat Fakültesi Dergisi. 2026;23(3):832-849. doi:10.33462/jotaf.1642095
BibTeX
@article{Altkat2026,
author = {Altıkat, Sefa and Gülsoy, Ersin and Kuş, Emrah and Gulbe, Alper and Altıkat, Alperay},
title = {Comparative Modelling of Walnut Kernel Weight Using Non-Destructive Physical and Mechanical Measurements},
journal = {Tekirdağ Ziraat Fakültesi Dergisi},
year = {2026},
volume = {23},
number = {3},
pages = {832--849},
doi = {10.33462/jotaf.1642095}
}
EUVIMED