CONDIS++: Cotton Growth Stage Detection Dataset

Overview

CONDIS++ is a computer vision dataset designed for object detection of cotton plant structures and growth stages The dataset supports research in precision agriculture, crop monitoring, and automated harvesting particularly for systems based on deep learning object detection models.

The dataset was created to enable robust detection of cotton bolls, cotton flowers and early buds under realistic agricultural conditions capturing the variability present in real greenhouse orchards.

CONDIS++ was developed in the context of research on cotton growth-stage detection using deep neural networks, and was used to train and evaluate the COTONET object detection architecture, a modified YOLO-based model for cotton phenological analysis.

Dataset Characteristics

The dataset contains images of cotton plants acquired under diverse visual conditions ensuring representativeness of real cultivation environments.

Key characteristics include:

• Multiple cotton growth stages
• Different perspectives and distances
• Natural lighting variability
• Occlusions and partial visibility
• Variations in plant morphology

Images include examples such as:

• Close and distant views of cotton plants
• Clear and partially occluded cotton bolls
• Occlusions caused by leaves or other cotton capsules
• Different illumination conditions (sunny, cloudy, shadowed)
• Perpendicular and angled camera viewpoints

These variations aim to ensure that machine learning models trained with CONDIS++ can generalize well to real-world agricultural deployments.

Annotations

The dataset provides object detection annotations corresponding to different phenological stages. Annotations are provided for each relevant object in the images, following the YOLO standard format.

Intended Applications

CONDIS++ supports research and development in precision agriculture, agricultural robotics and crop monitoring and phenotyping. In general, the dataset is particularly useful for training systems that need to detect cotton maturity and assist selective harvesting or plantation monitoring.

Acknowledgements

The authors wish to thank the personnel at CEBAS-CSIC for their invaluable contribution in providing the images used to construct the CONDIS++ dataset. Furthermore, we extend our gratitude for their expert knowledge and insights regarding the biological life cycles, production, and physiology of cotton plants, which were fundamental to this research. You can find more information about their research lines in https://www.cebas.csic.es/

Funding: This work has been partially funded by the DEMETER 5.0 project PLEC2022-009289 funded by MCIU/AEI/10.13039/501100011033 and by the ”European Union NextGenerationEU/PRTR”; by project SDC007/25/000183 supported by "Departament d'Empresa i Treball, Generalitat de Catalunya" and "European Union NextGenerationEU/PRTR” and also by the European Union under the project ARISE (HORIZON-CL4-2023-DIGITAL-EMERGING-01-101135959).

Citation

If you use the CONDIS++ dataset in your research, please cite the associated publication:

González, G., Alenyà, G., Foix, S.  
COTONET: A Custom Cotton Detection Network Based on YOLO for Growth Stage Detection.

(Currently under review).

Licence

This dataset is distributed under the licence specified in this Zenodo record. Please review the licence terms before using the dataset in derivative works or commercial applications.