Modelo predictivo basado en deep learning para identificar la Mancha Angular (Xanthomonas fragariae) en la hoja de Fresa para el Municipio del cantón Saraguro

Strawberry cultivation is a common agricultural activity in the canton of Saraguro. However, this crop has been significantly affected by the bacterium Xanthomonas fragariae, the causative agent of Angular Leaf Spot. This disease can rapidly infect large quantities of plants, particularly in open-field cultivation, leading to reduced production quality and significant economic losses. The objective of this study was to implement a deep learning-based model to detect Angular Leaf Spot in strawberry leaves. The methodology adopted was based on the Cross-Industry Standard Process for Machine Learning with Quality Assurance (CRISP-ML(Q)), adapting five phases: data understanding, data preparation, modeling, evaluation, and implementation. During the first and second phases, images were collected from various sectors of the canton of Saraguro, including La Matara, Tablón, Puente Chico, and Cañicapac, where strawberry crops are cultivated in both orchards and greenhouses. Additionally, images available on Kaggle were incorporated, resulting in a dataset of 2987 images, including both healthy leaves and those affected by Angular Leaf Spot. In the third and fourth phases, deep learning-based classification and object detection models, namely YOLOv5, YOLOv7, YOLOv8, and Faster R-CNN, were implemented and compared. These models were trained exclusively on data from the "Angular Leaf Spot" class, achieving accuracies of 93%, 89%, 92%, and 82%, respectively, with YOLOv5 emerging as the most accurate model. Subsequently, by integrating a complementary dataset containing images of healthy leaves into the training process, a slight but significant improvement in YOLOv5's accuracy was achieved, reaching 94%. In the final phase, a mobile application prototype was developed using the XP methodology and the Flutter framework to integrate the trained model. Final testing, conducted in collaboration with the Agricultural Development Technician of the municipal GAD, achieved an accuracy of 96% in a real-world environment. These results demonstrate that the proposed model is effective in detecting Angular Leaf Spot in strawberry leaves, providing a valuable tool for small-scale farmers in the canton of Saraguro Keywords: Angular leaf spot, CRISP-ML(Q), Deep learning, Yolov5

• Auteur principal: González Chillogalli, Diana Gabriela (author)
• Format: bachelorThesis
• Langue: spa
• Publié: 2025
• Sujets: .APRENDIZAJE PROFUNDO; CRISP–ML(Q); MANCHA ANGULAR; YOLOV5
• Accès en ligne: https://dspace.unl.edu.ec/jspui/handle/123456789/31847