Detection of Rot and Holes in Canchán Potatoes Using Artificial Vision and YOLOv8 Neural Networks

León León R.A.; Gutierrez Rodriguez J.A. · Memorias de la Conferencia Iberoamericana de Complejidad, Informatica y Cibernetica, CICIC (2025) · DOI: 10.54808/cicic2025.01.330

Resumen

Postharvest losses in Canchán potato, especially due to rot and holes, represent a critical problem for the industry, generating significant economic losses. Manual detection is inefficient and error-prone, requiring more accurate technological solutions. This project proposes an innovative machine vision system, based on the YOLOv8 architecture, to accurately detect rot and holes in Canchán potato tubers. The model was trained with 414 images of healthy and damaged potatoes, reaching an accuracy of 95% and an F1-score of 0.745 in independent tests, outperforming traditional methods. Images are captured with a smartphone under controlled lighting conditions and processed with the PIL and OpenCV2 libraries for analysis. The YOLOv8 network was trained using the Adam optimizer and data augmentation techniques to improve model generalization. This technological solution allows reducing post-harvest losses, improving product quality and optimizing marketing, contributing to the sustainability of potato production in Peru. Copyright 2025. © por el International Institute of Informatics and Systemics.

Instituciones: Universidad Privada del Norte