EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Computer Vision in Monitoring Fruit Browning: Neural Networks vs. Stochastic Modelling

Maria Kondoyanni iD, Dimitrios Loukatos iD, Charalampos Templalexis, Diamanto Lentzou iD, Georgios Xanthopoulos iD, Konstantinos G Arvanitis iD

Sensors · 15.04.2025 · 25 (8) · 2482

DOI: 10.3390/s25082482 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sensors, online erschienen am 15.04.2025. Verfasst von Maria Kondoyanni und 5 weiteren Autor:innen. Beteiligte Einrichtung: Department of Natural Resources Management and Agricultural Engineering, Agricultural University of Athens, 75 Iera Odos Str., 11855 Athens, Greece. Hauptthemen laut MEDLINE: Neural Networks, Computer, Fruit, Pyrus. Die Arbeit wurde bisher 10-mal zitiert (Crossref) und nennt 45 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

As human labour is limited and therefore expensive, computer vision has emerged as a solution with encouraging results for monitoring and sorting tasks in the agrifood sector, where conventional methods for inspecting fruit browning that are generally subjective, time-consuming, and costly. Thus, this study investigated the application of computer vision techniques and various RGB cameras in the detection and classification of enzymatic browning in cut pears, comparing convolutional neural networks (CNNs) with stochastic modelling. More specifically, light is shed on the potential of CNN-based approaches for high-throughput and easily adapted applications and the potential of stochastic methods for precise, quantitative analyses. In particular, the developed CNN model was easily trained and achieved an accuracy of 96.6% and an F1-score greater than 0.96 during testing with real pear slices. On the other hand, stochastic modelling provided quantitative indices (i.e., the Browning Index (BI) and Yellowing Index (YI)) derived from the CIE Lab* colour model, thus offering accurate monitoring of enzymatic browning and related optical changes but it was less versatile as it required human expertise for implementation and tuning. Using both the BI and YI as input vectors in the NN Bayesian classifier increased the correct classification rate of control samples to 82.85% (4.6% increase) and to 89.81% (15% increase) for treated samples. Finally, a future need for a hybrid approach combining the strengths of both methods was identified, with improved robustness and practicality of image analysis systems in agricultural quality control to enable higher levels of automation in this area.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Neural Networks, ComputerFruitPyrusStochastic ProcessesHumansBayes TheoremColor

Stichwörter der Autor:innen

CNNagricultural automationscomputer visionfood production chainfood qualityfruit browningstochastic modelling

Autor:innen und Institutionen

  • Department of Natural Resources Management and Agricultural Engineering, Agricultural University of Athens, 75 Iera Odos Str., 11855 Athens, Greece.

ORCID-Profile: Maria Kondoyanni, Dimitrios Loukatos, Diamanto Lentzou, Georgios Xanthopoulos, Konstantinos G Arvanitis

Zitieren

APA

Kondoyanni, M., Loukatos, D., Templalexis, C., Lentzou, D., Xanthopoulos, G., & Arvanitis, K. G. (2025). Computer Vision in Monitoring Fruit Browning: Neural Networks vs. Stochastic Modelling. Sensors, 25(8), 2482. https://doi.org/10.3390/s25082482

Vancouver

Kondoyanni M, Loukatos D, Templalexis C, Lentzou D, Xanthopoulos G, Arvanitis KG. Computer Vision in Monitoring Fruit Browning: Neural Networks vs. Stochastic Modelling. Sensors. 2025;25(8):2482. doi:10.3390/s25082482

BibTeX
@article{Kondoyanni2025,
  author = {Kondoyanni, Maria and Loukatos, Dimitrios and Templalexis, Charalampos and Lentzou, Diamanto and Xanthopoulos, Georgios and Arvanitis, Konstantinos G},
  title = {Computer Vision in Monitoring Fruit Browning: Neural Networks vs. Stochastic Modelling},
  journal = {Sensors},
  year = {2025},
  volume = {25},
  number = {8},
  pages = {2482},
  doi = {10.3390/s25082482}
}
Kostenlos starten →