Project-AgML/larch_casebearer_uav_detection

Larch Casebearer Uav Detection

Larch casebearer damage detection: 835 drone images (1500 x 1500 pixels) of larch forest in five areas of Vastergotland, Sweden (Bebehojd, Ekbacka, Jallasvag, Kampe, Nordkap), flown on May 27 and August 19, 2019, with 56,412 tree-crown bounding boxes labelled by damage from the larch casebearer moth: light_damage (33,298), other_tree (non-larch species, 11,391), high_damage (9,084), healthy_larch (2,638) and a single 'unknown' box. The imagery and labels were released by the Swedish Forest Agency (Forest Damages - Larch Casebearer); this copy is the YOLO-format Kaggle conversion (CDLA-Permissive 1.0) used by the Remote Sensing study DROMAL-Net, keeping its train (501), val (167) and test (167) split in a 'split' column and the survey area code (B01-B05) in 'site'. The dataset contains 835 images with 56,412 bounding box annotations across 5 categories.

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

@article{wang2026dromal,
  title={DROMAL-Net: A Forest Larch Casebearer Detection Model Focusing on Global Feature Enhancement and Positional Dynamic Clustering},
  author={Wang, Jiaxuan and Liu, Lu and Tang, Yuyang and Ma, Tao and Song, Xiangyu and Qiao, Pan and Ding, Zhen},
  journal={Remote Sensing},
  volume={18},
  pages={2916},
  year={2026},
  publisher={MDPI}
}

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