Multi-Temporal UAV Vineyard Segmentation Dataset — Riseholme (COCO Format, 2024–2025)

RGB images taken from a UAV of the Riseholme vineyard in Lincoln, UK, collected across three seasons: August 2024, March 2025, and July 2025. Images are annotated in COCO Segmentation format with four instance classes: pole, trunk, vine_row, and vineyard (canopy).

The dataset comprises three temporal subsets, each provided with train / validation / test splits:

Subset
Images
Annotations

Riseholme August 2024 (full resolution)
285
11,285

Riseholme March 2025 (full resolution)
287
18,199

Riseholme July 2025 (full resolution)
283
10,731

Total
855
40,215

Images were auto-oriented (EXIF stripping applied) with no augmentation. The dataset was used to train and evaluate YOLOv11 instance segmentation models for vineyard structure detection, supporting an integrated aerial-ground system for vineyard mapping and robotic row-level navigation.

A related dataset in YOLOv11 format (covering multiple UK vineyards at various altitudes) is available at: https://doi.org/10.5281/zenodo.15211733

If this dataset is helpful, please cite the associated journal paper:[to be filled in once published]