Dama12/yolov10x-onion-disease-detection

🧅 OnionScan AI : Automated Detection of Onion Diseases

This model is the core Computer Vision engine behind OnionScan, an intelligent mobile application designed for the automated detection of onion diseases in Burkina Faso. It is based on the state-of-the-art YOLOv10x architecture, fine-tuned specifically for agricultural diagnosis.

🎓 Academic Context

This project was developed as a Bachelor’s thesis at the Burkina Institute of Technology.

• Thesis Title: “Development of an Intelligent Mobile Application for the Automated Detection of Onion Diseases Using Computer Vision: Case of Burkina Faso.”
• Supervision: Dr. Rachid Gaetan NABOLLE
• Grade: 19/20 (4.0/4.0) – Excellent Distinction
• Impact: Demonstrated the practical use of Artificial Intelligence for sustainable agriculture in Burkina Faso.

📱 The OnionScan Application

While this repository hosts the AI model weights, the full solution is a comprehensive mobile application designed to adapt to the realities of rural environments (usable by farmers with limited technical skills).

Key Features of the App:

• Multi-source Input: Capture images directly via camera, import from the gallery, or use aerial images taken by drones (e.g., DJI Phantom) for large-scale fields.
• Offline / Online Mode: Fully functional in rural areas with low internet connectivity.
• Actionable Insights: Provides tailored treatment recommendations based on the AI diagnosis.
• Community & Tracking: Features a scan history, a community forum for farmer interaction, local notifications, and a built-in help center.

🧠 Model Details

• Architecture: YOLOv10-X (Extra-large for maximum precision)
• Task: Object Detection (Bounding boxes)
• Classes (4 Onion Diseases):
  1. alternariose (Alternaria porri)
  2. fusariose (Fusarium)
  3. pourriture (Rot)
  4. virose (Viral diseases)

📊 Training Metrics & Performance

The model was trained over 150 epochs (~47 minutes of training time) with carefully tuned hyperparameters (including Mosaic, Mixup, and Class Weights to handle dataset imbalance).

Key Validation Metrics:

• mAP50: 58.98 %
• mAP50-95: 36.88 %
• Precision: 59.63 %
• Recall: 54.99 %

(Detailed graphs, F1 curves, and the confusion matrix are available in the Files and versions tab of this repository).

💻 How to use the model (Inference)

You can test the model directly in Python using the ultralytics library:

!pip install git+https://github.com/THU-MIG/yolov10.git
from huggingface_hub import hf_hub_download
from ultralytics import YOLOv10

# Download the best weights from Hugging Face
model_path = hf_hub_download(repo_id="Dama12/yolov10x-onion-disease-detection", filename="weights/best.pt")

# Load the YOLOv10x model
model = YOLOv10(model_path)

# Run inference on an image
results = model.predict(source="path/to/your/onion_leaf.jpg", conf=0.25)
results[0].show()

⚖️ License & Citation

License

This model is built upon the YOLOv10 architecture. Therefore, the model weights are released under the AGPL-3.0 License (free for academic and research purposes). However, the dataset, methodology, and this specific implementation are intended strictly for non-commercial research purposes. If you intend to use this model for commercial applications, please contact the author.

Citation

If you use this model or the OnionScan application concept in your research, please cite it as follows:

@misc{dama2025onionscan,
  author = {Soumana Dama},
  title = {Development of an Intelligent Mobile Application for the Automated Detection of Onion Diseases Using Computer Vision: Case of Burkina Faso},
  year = {2024},
  publisher = {Burkina Institute of Technology},
  note = {Bachelor's Thesis, supervised by Dr. Rachid Gaetan NABOLLE}
}

👨‍💻 Developer Information

• Developed by: Soumana Dama — Full stack developer & AI engineer - Founder & Lead AI Engineer at Scoinvestigator AI
• LinkedIn: Soumana Dama
• GitHub: Damasoumana1

---

Built to empower sustainable agriculture through Artificial Intelligence. 🌾