Automated Cacao Disease Diagnosis Using Convolutional Neural Networks: A Roboflow-Powered Approach

Abstract

Cacao is a crop of vital economic and social impor- tance in Honduras, and its production faces significant challenges due to various diseases. This research presents a convolutional neural network (CNN)-based system aimed at improving the accuracy of disease detection in cacao fruit. A dataset of over 1,500 images of cacao fruits, both healthy and diseased, including conditions such as Moniliasis and Phytophthora, was collected, preprocessed, and labeled to train a CNN model using Roboflow. The results achieved include a mean Average Precision (mAP) of 90.50%, an accuracy of 88.30%, and a recall of 85.40%. These outcomes demonstrate that CNNs are essential for providing more accurate monitoring and better control over the health of cacao crops. The methodology incorporated advanced image preprocessing techniques and the implementation of a deep learning architecture for disease classification and detection.