Email: zijian.wang@uq.edu.au

Deep learning methods for image processing are rapidly advancing and imaging techniques have become a standard for classification and quantification in agriculture. Several studies already tackled crop segmentation of wheat under field conditions (Anderegg et al., 2023; Madec et al.,2023). However, most of the existing works focus on a two-classes segmentation problem, i.e., pixel-level classification of vegetation and background.

Global wheat (http://www.global-wheat.com/) aims to improve such efforts to train robust algorithms, which can segment leaves, stems, and inflorescences in complex wheat canopies. We collected ~40000 images from our phenotyping platforms spread across the globe.

The image information includes geographic location, developmental stage, genotype, or agricultural treatment. A diverse core set of 200 images is already labelled for all organs. Based on image tags and available meta-information, we currently select ~1000 images for further labelling.

This dataset will serve as a public benchmark for the training and validation of deep learning models. At IWC2024, we will i) present the dataset as a state-of-the-art benchmark for organ segmentation of wheat and ii) demonstrate the robustness of derived segmentation models under different testing environments.

Reference:

Anderegg, J., Zenkl, R., Walter, A., Hund, A., McDonald, B.A., 2023. Combining High-Resolution Imaging, Deep Learning, and Dynamic Modeling to Separate Disease and Senescence in Wheat Canopies. Plant Phenomics 5, 0053.

Madec, S., Irfan, K., Velumani, K., Baret, F., David, E., Daubige, G., ... & Weiss, M. 2023. VegAnn, Vegetation Annotation of multi-crop RGB images acquired under diverse conditions for segmentation. Scientific Data, 10(1), 302.