JoQert/argon-bilig-olive-dataset

Read in: 🇬🇧 English | 🇹🇷 Türkçe

🫒 Argon-Bilig: Background-Isolated Reference Disease Dataset for Gemlik Olives

Argon-Bilig is a standardized, high-resolution benchmark image dataset specifically engineered for the Gemlik olive cultivar (Olea europaea cv. 'Gemlik') across the Marmara basin. It is tailored to detect Olive Peacock Spot (Venturia oleaginea) and Olive Rust Mite (Aculus olearius) deformations while eliminating background environmental noise through custom YOLOv8 instance segmentation.

This dataset serves as the official benchmark repository for the Argon-Bilig Diagnostic Engine, developed under the TÜBİTAK 2204-A High School Research Projects Competition and the TEKNOFEST Agricultural Technologies Competition.

---

👥 Authors & Ownership

• Muhammed Emin DUYAR
• Ahmet Buğra GİRGİN
• Year: 2026

---

📦 Companion Pretrained Models

The official trained ONNX weights (YOLOv8s-seg & EfficientNet-B0 hybrid pipeline) evaluated on this dataset are available here:

• 🔗 Model Hub: JoQert/argon-bilig-olive-models

---

🔬 Data Engineering & Preprocessing Methodology

Raw agricultural imagery from fields and public repositories typically contains severe visual noise (e.g., human hands, soil, wooden tables, sky, uneven shadows). Such noise often induces deep learning models into shortcut learning rather than learning real pathological leaf patterns.

The Argon-Bilig Pipeline Steps:

1. Instance Segmentation: Raw images were processed via a custom-trained YOLOv8s-seg network to segment leaf and lesion boundaries at the pixel level.
2. Morphological Filtering: Leaf boundaries and lesion polygons were merged. Morphological closing (cv2.MORPH_CLOSE) and hole-filling contour algorithms were applied to repair pixel cracks and torn areas.
3. Background Isolation: All non-leaf background pixels were masked with pitch black RGB(0, 0, 0).
4. Normalized Crop: The isolated leaf was cropped with a relative margin (padding) around the bounding box.

---

🏷️ Dataset Structure & Class Distribution

The dataset contains a total of 3,447 images split using an 80/20 stratified scheme into training and validation sets:

argon-bilig-olive-dataset/
├── train/
│   ├── Healthy/                 # Healthy Olive Leaves (1,033 images)
│   ├── acerculus_olearius/      # Olive Rust Mite Deformations (398 images) *
│   └── olive_peacock_spot/      # Peacock Spot Lesions (1,325 images)
├── val/
│   ├── Healthy/                 # Healthy Olive Leaves (259 images)
│   ├── acerculus_olearius/      # Olive Rust Mite Deformations (100 images) *
│   └── olive_peacock_spot/      # Peacock Spot Lesions (332 images)
└── Cleaned_Class_Model_Dataset.zip  # Single-click full download archive

📌 Taxonomy Note on Class Labels:
The directory label acerculus_olearius represents the pest Olive Rust Mite (Aculus olearius, Family: Eriophyidae). The directory name is retained as acerculus_olearius for programmatic compatibility with trained inference weights.

---

💻 Quickstart with PyTorch

You can load and parse the dataset using standard PyTorch ImageFolder:

from torchvision import datasets, transforms
from torch.utils.data import DataLoader

# Standard ImageNet normalization pipeline
data_transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

# Load train split
train_dataset = datasets.ImageFolder('train', transform=data_transform)
train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)

print(f"Total training samples: {len(train_dataset)}")
print(f"Detected classes: {train_dataset.classes}")

---

📜 License & Academic Citation

This curated and preprocessed dataset is distributed under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.

If you use this dataset in your research or project, please cite:

@misc{argonbilig_dataset2026,
  title={Argon-Bilig: Background-Isolated Reference Disease Dataset for Gemlik Olives},
  author={Duyar, Muhammed Emin and Girgin, Ahmet Bu{\u{g}}ra},
  year={2026},
  howpublished={Hugging Face Datasets: \url{https://huggingface.co/datasets/JoQert/argon-bilig-olive-dataset}},
  note={Preprocessed and standardized olive leaf benchmark image repository}
}

---

🙏 Acknowledgements & Source Datasets

We gratefully acknowledge the researchers and open-access repositories whose raw images contributed to the foundation of this benchmark:

• Olive Peacock Eye Dataset: Diker, A., Elen, A., Közkurt, C. et al. (2024). Eur Food Res Technol, 250, 287–299. DOI: 10.1007/s00217-023-04386-8
• Olive Leaf Disease Unified Dataset: Abeer Saleh (CC0: Public Domain)
• Olive Leaf Image Dataset: Habibul Basher (CC0: Public Domain)