---
dataset_info:
  features:
  - name: bands
    dtype: binary
  - name: bands_shape
    list: int64
  - name: bands_dtype
    dtype: string
  - name: band_metadata
    dtype: string
  - name: band_order
    list: string
  - name: mask
    dtype: image
  - name: rgb
    dtype: image
  - name: split
    dtype: string
  splits:
  - name: train
    num_bytes: 351984049
    num_examples: 595
  download_size: 328624419
  dataset_size: 351984049
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: cc-by-4.0
task_categories:
- image-segmentation
size_categories:
- n<1K
---
# Tree Crown Segmentation

This dataset provides real multispectral imagery of Camellia oleifera tree crowns collected in a field environment in China. Captured using a UAV platform with a DJI Mavic 3 M drone featuring RGB and multispectral sensors, it offers high-resolution aerial data for semantic segmentation applications in agricultural monitoring. The dataset contains 595 images with pixel-level mask annotations.

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

The original train/test/val split has been preserved in the `split` column.

## Citation

```bibtex
@article{peng2026uav,
  title={UAV-based two-stage deep learning for tree crown segmentation and height estimation in Camellia oleifera plantations},
  author={Peng, Yongkang and Yan, Enping and Xu, Xiaocheng and Mo, Dengkui and Wei, Wei},
  journal={Ecological Informatics},
  volume={96},
  pages={103868},
  year={2026},
  publisher={Elsevier}
}
```

peng, . yongkang . (2026). experimental data [Figure]. Zenodo. https://doi.org/10.5281/zenodo.18505981


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