---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: bbch_stage
    dtype: string
  - name: label
    dtype:
      class_label:
        names:
          '0': Amaranthus_retroflexus (AMARE)
          '1': Amaranthus_tuberculatus (AMATU)
          '2': Chenopodium_album (CHEAL)
          '3': Echinochloa_crus-galli (ECHCG)
          '4': Setaria_faberi (SETFA)
  splits:
  - name: train
    num_bytes: 377491661
    num_examples: 3920
  download_size: 379048807
  dataset_size: 377491661
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: cc-by-4.0
task_categories:
- image-classification
size_categories:
- 1K<n<10K
---
# Wpdv2 Bbch Classification

This dataset provides real-world RGB images of agricultural weeds annotated using the BBCH growth stage classification system. It captures diverse weed species under typical field conditions relevant to precision agriculture applications. The images serve as a foundation for developing computer vision models focused on weed identification in crop environments. The dataset contains 3,920 images across 5 classes: Amaranthus_retroflexus (AMARE), Amaranthus_tuberculatus (AMATU), Chenopodium_album (CHEAL), Echinochloa_crus-galli (ECHCG), Setaria_faberi (SETFA).  
Images per class:
- Amaranthus_retroflexus (AMARE): 934
- Amaranthus_tuberculatus (AMATU): 409
- Chenopodium_album (CHEAL): 832
- Echinochloa_crus-galli (ECHCG): 768
- Setaria_faberi (SETFA): 977

This dataset is indexed on https://project-agml.github.io/ as part of the AgML python library.

## Citation

```bibtex
@article{fontaine2026taxonomical,
  title={Taxonomical loss for weed seedlings image classification},
  author={Fontaine, Hans-Olivier and Foucher, Samuel and Fallon, Edith and Simard, Marie-Jos{\'e}e and Lord, Etienne},
  journal={Scientific Reports},
  volume={16},
  number={1},
  pages={3837},
  year={2026},
  publisher={Nature Publishing Group UK London}
}
```

https://github.com/etiennelord/TaxonomicalLoss

*This dataset was reformatted from its original format to match HuggingFace standards.*