PubMed · 2026-05-21
Researchers built a large, labeled dataset of kiwifruit vine photos and GPS-tagged videos to train computers to automatically track which growth stage each plant is in — from dormant bud through fruit set — replacing slow, expensive manual scouting across orchards.
The dataset contains 1,665 annotated images covering phenological stages from bud break through fruit set, organized by plant structure, gender, and growth stage.
A custom 17-class growth-stage system was developed by adapting the standard BBCH scale — merging visually similar stages and removing categories that were too ambiguous to label reliably.
Georeferenced (GPS-tagged) videos with manual ground-truth maps were included specifically to validate automated counting algorithms at the spatial level, not just image recognition accuracy.
