Project-AgML/wild_rice_kernel_quality_classification

Wild Rice Kernel Quality Classification

RGB images of single Canadian wild rice (Zizania palustris) kernels photographed under controlled laboratory conditions at the University of Manitoba with a Canon EOS R100 and cropped to 512x512 patches. Kernels are graded into eight quality classes: Healthy-Large, Healthy-Medium, Discoloured-Low, Discoloured-High, Broken-Low, Broken-High, Insect-damaged and Unhulled. The raw config holds the 16,632 original images and the augmented config the record's 49,896 augmented copies (three augmentation sets, A to C, per original). The paired VNIR hyperspectral data are in a separate record. The dataset contains raw and augmented versions.
The raw dataset contains 16,632 images.
Images per class:

• broken_high: 2,124
• broken_low: 2,124
• discoloured_high: 2,052
• discoloured_low: 2,016
• healthy_large: 2,088
• healthy_medium: 2,052
• insect_infected: 2,160
• unhulled: 2,016

The augmented dataset contains 49,896 images.
Images per class:

• broken_high: 6,372
• broken_low: 6,372
• discoloured_high: 6,156
• discoloured_low: 6,048
• healthy_large: 6,264
• healthy_medium: 6,156
• insect_infected: 6,480
• unhulled: 6,048

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

Citation

@article{sikiru2026multimodal,
  title={A multimodal imaging dataset for quality grading of Canadian wild rice kernels using RGB and VNIR hyperspectral data},
  author={Sikiru, Yinka and Erkinbaev, Chyngyz},
  journal={Data in Brief},
  volume={65},
  pages={112601},
  year={2026},
  publisher={Elsevier}
}

The dataset itself can be cited as:

Sikiru, Y., & Erkinbaev, C. (2026). RGB dataset of canadian wild rice kernels [Dataset]. Zenodo. https://doi.org/10.5281/ZENODO.18346354

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