Bark_Merged

Description

This dataset is a unified bark-image collection for tree species classification, built by merging three publicly available bark datasets into a single, consistently labeled benchmark. It contains 32,078 labeled images spanning 66 distinct tree species, captured across different geographic regions to improve species diversity and model generalization. The merged dataset integrates images from three sources: BarkNet 1.0 Images: 23,000 Classes: 23 BarkVN-50 Images: 5,578 Classes: 50 Wood Species (Bangalore) Images: 3,500 Classes: 22 Merged: Images 32,078 Classes:66 Images relabeled and reorganized into 66 consistent classes (labels 0–65). Only classes with approximately 100–300 images retained for balance. All images resized to 224 × 224 pixels for standard CNN input. Suggested split: 70% training / 15% validation / 15% testing. If you use this dataset please cite the work: @inproceedings{ali2025bark, author = {Ali, Aroshi and Jahan, Nusrat and Sagor, Anite Halim and Alam, S. M. Jahangir and Ahmed, Saad}, title = {Lightweight and Interpretable {CNNs} for Bark-Based Tree Species Classification}, booktitle = {2025 28th International Conference on Computer and Information Technology (ICCIT)}, year = {2025}, address = {Cox's Bazar, Bangladesh}, doi = {https://doi.org/10.1109/ICCIT68739.2025.11490516} }

Institutions

• Bangladesh Army University of Science and Technology

  Rangpur Division, Saidpur

Categories

Licence