Skip to content
Open Data Finder
GovData Germany

Annotated image dataset for weed detection in cereals

Published by OpenAgrar

About this dataset

Artificial intelligence (AI) offers great potential for the detection and classification of weeds in arable farming. A key prerequisite for the development of high-performance AI models is the availability of large, high-quality and precisely annotated image datasets. The weed image dataset presented here contributes to this. The image data presented here were collected in wheat fields at 12 trial sites using a Sony Alpha 6000 between 2021 and 2023 in the Braunschweig area (Germany). Field surveys served as ground truth values and a reference for the subsequent annotation of the images in COCO 1.0 format using CVAT. Each plot was assessed by two herbology experts. The images were then annotated as ellipses by two experts and cross-checked for quality in accordance with the dual-review principle. The 55 annotation classes cover 54 EPPO codes and “Unknown”, which includes weeds that could not be reliably identified either during the field assessment or during annotation. In total, the dataset provides 1,480 high-resolution RGB images with 20,589 annotations, which can contribute to the development of advanced AI models for weed detection. These serve as the cornerstone for efficient agricultural land management by reducing the use of plant protection products and preserving crop biodiversity in precision farming.

Annotated image dataset for weed detection in cereals · Open Data Finder