Detection of green tea buds and spatial mapping using low-altitude UAV imagery and object detection models

Accurate collection of Camellia sinensis buds is essential for producing high-quality green tea; however, their small size, color similarity to surrounding leaves, and frequent occlusion make automated detection and yield estimation difficult. Existing UAV-based yield mapping has been applied to several crops, yet its application to green tea using bud-level indicators remains largely unexplored. This study develops a low-altitude UAV and deep-learning system for detecting field-grown tea buds and generating spatial yield distribution maps in commercial plantations. YOLOv11 and YOLOv12 models with their nano, small, and large variants were used in this study to detect tea buds from UAV imagery. Among them, YOLOv11-L achieved the highest performance, with a detection accuracy of 83.95% mAP@0.5 and demonstrated a strong correlation with harvested yield (R² = 0.72) with Root Mean Square Error (RMSE) of 50.999 and a relative RMSE (rRMSE) of 26.02% using low-altitude (5 m) UAV imagery. Georeferenced detections were used to generate spatial distribution maps identifying high- and low-yield zones. The proposed UAV–YOLO framework demonstrates the feasibility of bud-based spatial yield estimation and is suitable for targeted harvesting and precision management in green tea production.