Litchi-UAV: A UAV-based Litchi Fruit Detection Dataset for Precision Agriculture

This dataset contains UAV-acquired litchi fruit images collected from an orchard in Conghua District, Guangzhou, China, on two dates: May 14, 2024 (cloudy to overcast) and July 2, 2024 (sunny). Images were captured using a DJI Phantom 4 UAV with both vertical and oblique shooting methods, at a resolution of 4096×2160 pixels.

The dataset is intended for training and evaluating object detection models, particularly lightweight architectures suitable for edge deployment. It includes 432 original high-resolution images, which were preprocessed by sliding‑window cropping into 1024×1024 patches and augmented with Gaussian noise, salt‑and‑pepper noise, brightness adjustments, and darkening to improve model robustness. The final dataset consists of 1756 training images, 86 validation images, and 44 test images, accompanied by YOLO‑format bounding‑box annotations for litchi fruits. Various occlusion types (fruit, branch, leaf) are represented in the images.

This dataset supports the research described in the paper "Edge‑YOLOv11: Lightweight UAV Fruit Detection in Dense Canopies for Precision Agriculture" and is made publicly available to facilitate reproducible research in precision agriculture and UAV‑based crop monitoring.