Creación de un modelo para la clasificación de imágenes de la mancha bacteriana en la hoja de tomate utilizando el modelo VGG16

Advancements in artificial intelligence have revolutionized sectors such as agriculture, highlighting its ability to optimize processes through automated data analysis. In this context, convolutional neural networks have proven to be fundamental tools for image processing and classification, enabling the efficient identification of complex patterns. These technologies have been applied in early disease detection, crop growth monitoring, and quality assessment of agricultural products, providing innovative solutions. This curricular integration project focused on designing, tuning, and evaluating a deep learning model based on the VGG16 architecture for the automated classification of tomato leaf images affected by bacterial spots. The methodology used was CRISP-ML(Q), structured into three main phases:Data engineering, where images were identified and collected from the Kaggle repository. These images were processed using cleaning, balancing, and data augmentation techniques, resulting in a final dataset of 3,816 images, increasing both diversity and quality.Model engineering, where the pre-trained VGG16 architecture on ImageNet was implemented, optimizing hyperparameters and adjusting classification layers to fit the required task.Model evaluation, where training was conducted on the Google Colab platform, yielding a model with optimal performance. The final evaluation showed an overall accuracy of 99.73% and a minimal loss of 0.78%, surpassing previous studies with similar configurations. Additionally, the zero-shot learning technique was applied to assess the model’s generalization ability using data collected from real-world environments, achieving an accuracy of 75%. The developed model optimizes pesticide use, enhances productivity, and promotes agricultural sustainability. Therefore, it is recommended to expand the dataset with images obtained in real-world environments and explore its integration with mobile technologies.

• Autor principal: Quizhpe Hurtado, Jennifer Elizabeth (author)
• Format: bachelorThesis
• Idioma: spa
• Publicat: 2025
• Matèries: PRECISON GLOBAL; REDES NEURONALES CONVOLUCIONALES; MANCHA BACTERIANA
• Accés en línia: https://dspace.unl.edu.ec/jspui/handle/123456789/32171