Detection and Quantification of Citrus Unshiu Mandarins with YOLO: A Comparative Analysis

Isidro E.; Corbera M.; Jimenez G.; Romero S. · Proceedings of the 2025 IEEE 32nd International Conference on Electronics, Electrical Engineering and Computing, INTERCON 2025 (2025) · DOI: 10.1109/intercon67304.2025.11244684

Resumen

This study presents deep-learning-based methods to identify and quantify Citrus unshiu (Satsuma mandarin) using machine learning models applied to agricultural imagery. Citrus unshiu is a species of high commercial value due to its early-season harvest periods and adaptability to various climates. In Peru, the national production of mandarins reached 647,800 metric tons in 2024, with the majority concentrated in Lima and Ica. However, only a fraction is destined for export, despite the growth in international demand. For this purpose, we adopt YOLO (You Only Look Once) object detection architecture. YOLO's one-stage design makes it suitable for tasks that require fast processing of high resolution images, like precision agriculture technologies. The results suggest a reliable detection system, suitable for mass implementation. It also shows that recent state of the art architectures and techniques improved the segmentation analysis of previous YOLO models.

Instituciones: Pontificia Universidad Católica del Perú