Diseño de un modelo de Inteligencia artificial en la detección de Colletotrichum. spp y Oídium . spp en el cultivo de taxo (Passiflora mollisima B.H.K

Abstract

In this work, an Artificial Intelligence model based on the YOLOv11 architecture was developed for the segmentation and detection of three relevant classes in taxum (Passiflora mollisima B.H.K) culture: Colletotrichum spp., Oidium spp. and healthy leaves, in addition to the background class. A total of 1054 tagged images were used, of which 37 were used for testing and 40 for validation. The training took place over 60 eras. The results showed moderate precision (0.4994), recall (0.46815) and mAP@50 (0.38459) values, reflecting that the model manages to correctly identify about half of the positive instances, but presents confusions between classes with similar visual characteristics (in particular, healthy leaves and the background category). The confounding matrix corroborates these findings, evidencing the need to increase the diversity and quantity of training data, as well as to refine the model's hyperparameters to improve class discrimination. Despite the limitations, progressive decreases in loss curves during training and validation indicate a stable learning process. It is concluded that the model can perform a segmentation and initial detection of fungal diseases and healthy leaves, offering a valuable starting point for future improvements that contribute to a more efficient precision agriculture in taxo cultivation.