Datasets:

Project-AgML
/

wheat_seed_variety_classification

dataset_info:
  features:
    - name: image
      dtype: image
    - name: label
      dtype:
        class_label:
          names:
            '0': Akbar-19
            '1': Dilkash-20
            '2': Urooj-22
    - name: seed_id
      dtype: string
  splits:
    - name: train
      num_bytes: 31096678
      num_examples: 1124
  download_size: 31104631
  dataset_size: 31096678
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
task_categories:
  - image-classification
size_categories:
  - 1K<n<10K

Wheat Seed Variety Classification

Single wheat seeds of three Pakistani varieties, Akbar-19, Dilkash-20 and Urooj-22, photographed at the University of Agriculture, Faisalabad, for varietal purity identification. After checking seed morphology under a microscope, the authors selected 125 seeds per variety and photographed each from the back, front and side with a Canon EOS 5D Mark III; this release holds the images resized to 256 x 256 (1,124 images). The label is the variety and seed_id groups the views of the same seed (A1-A125, D1-D125, U1-U125); which view each image shows is not recorded in the file names. The dataset contains 1,124 images across 3 classes: Akbar-19, Dilkash-20, Urooj-22.
Images per class:

Akbar-19: 374
Dilkash-20: 375
Urooj-22: 375

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

Citation

@article{nawaz2025high,
  title={High-resolution RGB image dataset for wheat seed varietal identification and purity assessment},
  author={Nawaz, Mehreen and Safder, Sadaf and Riaz, Shazia and Ali, Saqib},
  journal={Data in Brief},
  volume={61},
  pages={111690},
  year={2025},
  publisher={Elsevier}
}

The dataset itself can be cited as:

paal lab, precision lab. (2025). Varietal purity of wheat seeds dataset [Dataset]. Mendeley Data. https://doi.org/10.17632/W5248V9FK3.1

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