Dataset on UAV RGB videos acquired over a vineyard including bunch labels for object detection and tracking.

Abstract

Counting the number of grape bunches at an early stage of development offers relevant information to the winegrower about the potential yield to be harvested. However, manual counting on the fields is laborious and time-consuming. Remote sensing, and more precisely unmanned aerial vehicles mounted with RGB or multispectral cameras, facilitate this task rapidly and accurately. This dataset contains 40 RGB videos from a 1.06-ha vineyard located in northern Spain. Moreover, the dataset includes mask labels of visible grape bunches. The videos were acquired throughout four UAV flights with an RGB camera tilted at 60 degrees. Each flight recorded one side of a row of the vineyard. The grape berries were between pea-size (BBCH75) and bunch closure (BBCH79) stage, which is two months before harvesting. No operations other than those usual in a commercial vineyard, such as pruning, cane tying, fertilization, and pest treatment, have been carried out, hence, the dataset presents leaf occlusion. The dataset was gathered and labelled to train object detection and tracking algorithms for grape bunch counting. Furthermore, it eases the work of winegrowers to check the sanitary status of the vineyard.

Plant phenotyping relevance

ブドウ房数という植物の収量関連形質をUAV画像から推定するための、ラベル付き動画データセットであり、物体検出・追跡手法の開発を支援する中心的な成果である。

Code and data availability

This Data in Brief article describes its own public dataset: 40 UAV RGB videos over a vineyard with grape bunch mask annotations (MOTS-style PNG labels) for object detection/tracking and phenotyping, deposited on Zenodo with an explicit direct URL and DOI. This is a paper-specific, publicly available, directly reproduc