MedicinalLeafBD v1.0: A Leakage-Safe Medicinal Plant Leaf Dataset from Bangladesh

MedicinalLeafBD v1.0 is a curated and leakage-safe medicinal plant leaf dataset developed by Ovi Sarker to support deep learning research in plant recognition and medicinal botany. The dataset contains 5,454 high-quality images from 16 medicinal plant species commonly found in Bangladesh. Each image was standardized to 224×224 pixels and organized using a leakage-safe split strategy (80% training, 10% validation, 10% testing) to prevent data leakage between sets. Duplicate and near-duplicate images were removed using perceptual hashing (pHash). Classes include: Ashok, Basil, Cannonball Tree, Cordia, Debdaru, Heaven Lotus, Jarul, Mastwood, Minjiri, Nageshor, Neem, Punnag, Royna, Telakucha, Thankuni, and Vasaka. All images were collected from publicly available open-access online resources (educational archives, botanical repositories, and Google Images). Data was then restructured, relabeled, and verified manually by the author. This dataset aims to assist researchers and students working in AI-based medicinal plant classification, transfer learning, and explainable AI (Grad-CAM) visualizations. **Key Features:**- 5,454 leaf images across 16 plant species- Leakage-safe 80/10/10 data split- Duplicate removal via perceptual hashing- Ready for TensorFlow/Keras transfer learning- Model metadata file (`model_meta.json`) included **Citation:**O. Sarker, "MedicinalLeafBD v1.0: A Leakage-Safe Medicinal Plant Leaf Dataset from Bangladesh," Zenodo, 2025. DOI: 10.5281/zenodo.17389084. **License:** Creative Commons Attribution 4.0 International (CC BY 4.0)**Contact:** [e-mail]