Evet-Africa/poultry-disease-detector

🐔 Poultry Disease Detector — YOLOv11n Fine-tuned

A fine-tuned YOLOv11n model for detecting 26 poultry disease lesions and clinical signs from images. Trained on a custom annotated dataset covering Newcastle Disease, Coccidiosis, and Fowlpox.

This model is intended to assist farmers, veterinary researchers, and agricultural AI developers in early detection of poultry diseases.

---

📊 Model Performance

Trained for 100 epochs on a custom dataset of ~400 annotated images.

Metric	Value
mAP@50	83.7%
mAP@50–95	66.4%
Precision	71.2%
Recall	73.7%

---

🏷️ Detectable Classes (26)

The model detects the following clinical signs and lesions:

#	Class	Associated Disease
0	Torticollis	Newcastle Disease
1	Oculonasal Discharge	Newcastle Disease
2	Facial Swelling	Newcastle Disease
3	Conjunctival Hemorrhage	Newcastle Disease
4	Tracheal Congestion	Newcastle Disease
5	Proventricular Hemorrhage	Newcastle Disease
6	Breast Muscle Congestion	Newcastle Disease
7	Pulmonary Congestion	Newcastle Disease
8	Intestinal Hemorrhage	Newcastle Disease
9	Bursal Edema	Newcastle Disease
10	Ovarian Follicle Congestion	Newcastle Disease
11	Intestinal Congestion	Coccidiosis
12	Intestinal Necrosis	Coccidiosis
13	Cecal Core	Coccidiosis
14	Enlarged Cecum	Coccidiosis
15	Pericardial Hemorrhage	Newcastle Disease
16	Cutaneous Nodules	Fowlpox
17	Periocular Hyperemia	Fowlpox
18	Oral & Pharyngeal Plaques	Fowlpox
19	Tracheal Fibrinous Exudate	Fowlpox
20	Splenic Congestion	Newcastle Disease
21	Scab Formation	Fowlpox
22	Ceca Hemorrhage	Coccidiosis
23	Liver Congestion	Newcastle Disease
24	Congested Lung	Newcastle Disease
25	Cyanotic Comb & Wattle	Newcastle Disease

---

🚀 Usage

Installation

pip install ultralytics

Inference

from ultralytics import YOLO

# Load the model
model = YOLO("model.pt")

# Run detection on an image
results = model("path/to/poultry_image.jpg")

# Show results
results[0].show()

# Print class predictions
for box in results[0].boxes:
    class_id = int(box.cls)
    confidence = float(box.conf)
    class_name = results[0].names[class_id]
    print(f"{class_name}: {confidence:.2%}")

Batch Inference

from ultralytics import YOLO

model = YOLO("model.pt")

# Run on a folder of images
results = model("path/to/images/", save=True)

---

📁 Repository Structure

├── model.pt                  # Fine-tuned model weights (best)
├── yolo11n.pt                # Base YOLOv11n pretrained weights
├── dataset.yaml              # Dataset configuration (26 classes)
├── clean.py                  # Dataset cleaning utility
├── images/                   # Labeled training images
├── labels/                   # YOLO-format annotation .txt files
├── unlabeled_images/         # Additional unannotated images
├── models/
│   ├── history_model.keras   # Companion history/feature model
│   ├── history_scaler.pkl    # Feature scaler
│   └── history_feature_vectors.csv
└── runs/detect/train/        # Training run outputs & metrics

---

🗂️ Dataset

• ~406 labeled images across 3 disease categories
• 26 lesion/sign classes annotated in YOLO format
• Annotations created using CVAT
• Images sourced from poultry farm and veterinary postmortem contexts

Disease Categories Covered

• Newcastle Disease — systemic, highly contagious viral disease
• Coccidiosis — intestinal parasitic infection
• Fowlpox — slow-spreading viral disease

---

🏋️ Training Configuration

model: yolo11n.pt
epochs: 100
batch: 4
imgsz: 640
optimizer: auto
patience: 20
pretrained: true
augment: true (randaugment, mosaic, fliplr)

---

⚠️ Limitations & Disclaimer

• This model is a research/assistive tool and should not replace veterinary diagnosis.
• Performance may vary on images taken in different lighting conditions or camera setups compared to training data.
• The dataset size (~400 images) is relatively small; predictions should be validated by a qualified veterinarian.
• Model was trained and validated on the same image set (train/val split from same distribution).

---

📜 License

This model is released under the AGPL-3.0 License, inherited from Ultralytics YOLOv11.

Any downstream use, modification, or distribution must also be open-sourced under AGPL-3.0. For commercial use without open-sourcing, contact Ultralytics for an Enterprise License.

---

🙏 Citation

If you use this model in your research, please cite:

@misc{poultry-disease-detector-2025,
  title        = {Poultry Disease Detector — Fine-tuned YOLOv11n},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/Evet-Africa/poultry-disease-detector}},
  note         = {Fine-tuned on custom CVAT-annotated poultry disease dataset}
}

@software{yolo11_ultralytics,
  author  = {Glenn Jocher and Jing Qiu},
  title   = {Ultralytics YOLO11},
  version = {11.0.0},
  year    = {2024},
  url     = {https://github.com/ultralytics/ultralytics},
  license = {AGPL-3.0}
}