Modelo basado en redes neuronales convolucionales (CNN) para la detección y clasificación de la moniliasis en el cacao.

Abstract

Moniliasis, caused by Moniliophthora roreri fungus, is one of the main diseases affecting cocoa production in Latin America, leading to significant losses. Early detection of this disease is crucial for implementing effective control strategies and minimizing economic impact. This article presents a model based on convolutional neural networks (CNN) for automatic detection and classification of moniliasis in cocoa pods using digital images. A total of 2,000 images of healthy and affected (moniliasis) cocoa were collected and preprocessed using data augmentation, data standardization, and resizing techniques. The model was trained with a sequential CNN architecture, including convolutional, pooling, flattening, and dense layers, optimized for binary classification. The model's performance was evaluated using precision, sensitivity, F1 score, and confusion matrix metrics, achieving a maximum accuracy of 98.3% in training and 99% in validation. The above results show that CNNs are effective tools for the automated identification of moniliasis, providing farmers with a technological resource that facilitates early detection and improves decision-making in crop management. This study demonstrates the great potential of artificial intelligence in precision agriculture, contributing to the sustainability and productivity of cocoa in Ecuador.