A Benchmark Dataset and Baseline Model for Disease Detection in Peruvian Cocoa Fruits Using YOLOv5

Leon G.; Yauri J. · Proceedings of 8th International Conference on Systems Engineering - Cybersecurity and AI: Building a reliable digital future, CIIS 2025 (2025) · DOI: 10.1145/3771678.3771682

Resumen

Cocoa production is crucial for the social and economic development of many families worldwide. Despite not ranking among the top cocoa producing countries globally, Peru is recognized for its high-quality product in international markets. The low productivity of cocoa is due to the lack of good agricultural practices and the increase of diseases and pests that affect it. This paper presents a new reference dataset that includes images of high resolution of diseases and pests on cocoa fruits or pods that were collected in the Valle del Río Apurímac, Ene, y Mantaro (VRAEM) region of Peru. The dataset contains images of five classes of cocoa fruits: Healthy, Phytophthora palmivora fungus, Moniliophthora roreri fungus, Moniliophthora perniciosa fungus, and Carmenta foraseminis pest. An early detection of diseases and pests in cocoa fruits, along with appropriate treatment, can help farmers mitigate their economic losses. In addition, a reference model for early detection of diseases in cocoa fruits using YOLOv5 is also provided, showing an excellent classification performance, with 97.06% sensitivity and 97.25% specificity in the binary classification of unhealthy versus healthy cocoa pods. Based on this baseline result, it can be concluded that the dataset is suitable for binary classification in real-world scenarios and can be used for future research.

Instituciones: Universidad Nacional de San Cristóbal de Huamanga