Project-AgML/medicinal_plant_parts_classification

Medicinal Plant Parts Classification

Medicinal plant part identification: 15,100 leaf photographs (224 x 224 pixels) of 52 traditional medicinal plant species from Ethiopia, taken with Samsung A30s and Tecno Spark K7 smartphone cameras under varying light and leaf orientation (mostly 300 images per species). Each leaf image is labelled with the plant part(s) used to prepare medicine from that species (leaf, seed, root, bark, root_leaf, leaf_seed, bark_leaf, or un_identified), as assigned by five traditional-medicine practitioners; the label is a property of the species, tagged onto its leaf image, rather than something visible in the photo. A 'species' column gives the local plant name from the file names (e.g. gesho, tenadam, rosmery, zeytuna). The source table's 'barck' spelling is normalized to 'bark'. The dataset contains 15,100 images across 8 classes: leaf, seed, root, bark, root_leaf, leaf_seed, bark_leaf, un_identified.
Images per class:

• leaf: 5,300
• seed: 2,400
• root: 1,800
• bark: 900
• root_leaf: 1,800
• leaf_seed: 900
• bark_leaf: 600
• un_identified: 1,400

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

Citation

@article{widneh2025identification,
  title={Identification of medicinal plant parts using depth-wise separable convolutional neural network},
  author={Widneh, Misganaw Aguate and Alemu, Amlakie Aschale},
  journal={PLOS One},
  volume={20},
  pages={e0322936},
  year={2025},
  publisher={Public Library of Science}
}

The dataset itself can be cited as:

Aguate, M. (2024). Medicinal Plant leaf images [Dataset]. figshare. https://doi.org/10.6084/M9.FIGSHARE.26426806.V1

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