Digital Twin and Augmented Reality Framework for Spatio-Temporal Risk Assessment of Anthracnose in Export Banana Clusters

Abstract

Postharvest fungal diseases represent one of the main factors affecting fruit quality and commercial value in global banana production systems. Among these pathogens, anthracnose caused by Colletotrichum musae is particularly challenging because infections often remain latent during early fruit development and only become visible during postharvest stages. This biological behavior complicates early detection and increases the risk of economic losses during storage and export logistics. This research proposes an integrated technological framework that combines environmental monitoring, machine learning prediction, digital twin simulation, and augmented reality visualization to estimate the spatio-temporal risk of anthracnose development in export banana clusters. Field monitoring was conducted between June and December 2025 in three commercial banana plantations located in Los Ríos, Ecuador, including Reybanpac, Ubesa, and Exportadora Bananera Noboa. A multimodal dataset named RIOSBAN-RA was constructed using continuous environmental monitoring, high-resolution RGB imagery, and three-dimensional scanning of banana clusters. The dataset contains 134,400 environmental observations, 400 high-resolution RGB images, and 50 three-dimensional cluster models. Correlation analysis revealed that prolonged exposure to relative humidity levels above 90% showed the strongest relationship with disease occurrence (r = 0.61, p < 0.001). Three predictive algorithms were evaluated: logistic regression, support vector machines, and random forest models. Among them, the random forest algorithm achieved the highest performance with an accuracy of 86.7% and an AUC value of 0.91. The predictive model was subsequently integrated into a digital twin environment developed in Unity 3D, enabling the simulation of disease risk under varying environmental conditions. Additionally, an augmented reality interface was implemented to visualize predicted infection zones directly on banana clusters during field inspections. Validation experiments involving agricultural technicians demonstrated improvements in both inspection efficiency and disease risk interpretation. The results suggest that integrating digital twins, machine learning models, and augmented reality visualization can significantly enhance early disease detection and decision-making processes in export banana production systems