Artificial vision model based on convolutional neural networks for black pod identification in cacao plantations

Villalobos-Culqui C.; García-Rivas-plata C.; Tuesta-Hidalgo O.A. · Revista Cientifica de Sistemas e Informatica (2025) · DOI: 10.51252/rcsi.v5i1.678

Resumen

Early detection of black pod in cocoa plantations represents a key challenge in the agricultural sector, as it affects yield and grain quality. The lack of advanced methods hinders timely identification. This study develops artificial vision models based on convolutional neural networks (CNN) to improve detection. Over nine months, we collected and labeled 1,982 images of affected pods from five plots in the Shitarillo sector, Alto Saposoa district, San Martín. We implemented YOLOv8, InceptionV3, and VGG19, applying transfer learning to optimize classification. The dataset was split into 70% for training, 20% for validation, and 10% for testing. YOLOv8 and InceptionV3 achieved an average accuracy of 79%, outperforming VGG19. Evaluation metrics, along with ANOVA and Tukey tests, confirmed that both models provided superior performance with no significant differences between them. YOLOv8 stood out for its greater robustness and accuracy, suggesting its implementation in early detection systems to optimize disease control in cocoa plantations. © The authors.

Instituciones: Universidad Nacional Autónoma de Alto Amazonas; Universidad Nacional Ciro Alegría; Universidad Nacional de San Martín