Chicken Disease Classification

CNN-based classifier to detect Coccidiosis vs Healthy from chicken fecal images. The project is production-lean with modular pipelines, configuration-driven training, DVC for reproducibility, and a minimal Flask app for serving predictions.

Highlights

• VGG16 backbone with configurable head (see params.yaml)
• Reproducible training using DVC stages (dvc.yaml)
• Clean component/pipeline layout under src/cnnClassifier
• Flask UI and JSON API (app.py, templates/, static/)
• Metrics tracked in scores.json

---

Quick Start

Prerequisites

• Python 3.8 or higher
• Git
• DVC (pip install dvc)

On Windows PowerShell:

# 1) Create and activate a virtual environment
python -m venv .venv
./.venv/Scripts/Activate.ps1

# 2) Install dependencies
python -m pip install --upgrade pip
pip install -r requirements.txt

# 3) Reproduce the full pipeline (data -> model)
dvc repro

# 4) Optional: visualize pipeline graph
dvc dag

Alternatively, you can run the full training pipeline directly:

python main.py

Artifacts (datasets, models, logs) are produced under the artifacts/ directory as configured in config/config.yaml.

---

Serve the Model (Flask)

Run the web app with UI and JSON endpoint:

python app.py

• UI: open http://127.0.0.1:8080/ and upload an image.
• API: POST /predict with JSON body { "image": "<base64>" }.

Minimal request example:

{
	"image": "<base64-encoded-image>"
}

Response shape:

[
	{ 
        "prediction": "Healthy | Coccidiosis", 
        "status": "success", 
        "message": "Prediction completed successfully" 
    },
	{ "image": "<echoed-input-as-base64>" }
]

---

Configuration

• Pipeline and I/O paths: config/config.yaml
• Hyperparameters: params.yaml
• DVC stages: dvc.yaml
• Logs: logs/running_logs.log

Model details (default):

• Base: VGG16 (Keras Applications)
• Input size: IMAGE_SIZE from params.yaml (default: 224x224x3)
• Top: Flatten + Dense(softmax, CLASSES)
• Optimizer: SGD(learning_rate = LEARNING_RATE)
• Training config: BATCH_SIZE, EPOCHS, AUGMENTATION

---

Project Structure

.
├── app.py                         # Flask app (UI + API)
├── main.py                        # Run full training pipeline sequentially
├── dvc.yaml                       # DVC stages (ingestion, base, train, eval)
├── params.yaml                    # Hyperparameters
├── config/config.yaml             # Paths and stage configs
├── src/cnnClassifier/             # Components, pipelines, utils
│   ├── components/                # data_ingestion, prepare_base_model, training, evaluation, callbacks
│   ├── config/                    # Configuration management
│   ├── constants/                 # Constant values (e.g., config and params file path)
│   ├── entity/                    # Data classes for configs
│   ├── pipeline/                  # stage_01..04 + predict.py
│   ├── utils/                     # common helpers (e.g., base64 decode)
│   └── ...
├── artifacts/                     # Data + models produced by stages
│   ├── data_ingestion/Chicken-fecal-images/{Coccidiosis,Healthy}/
│   ├── prepare_base_model/
│   └── training/model.h5
├── templates/index.html           # Frontend
├── static/style.css               # Styles
├── scores.json                    # Evaluation metrics (DVC-tracked)
├── requirements.txt
├── Dockerfile
└── README.md

---

Reproducible Training with DVC

Common commands:

# Run all stages
dvc repro

# Show pipeline graph
dvc dag

# Inspect metrics
dvc metrics show

Stages defined in dvc.yaml:

1. data_ingestion → 2) prepare_base_model → 3) training → 4) evaluation

You can also run the same stages via python main.py.

---

Docker

Build and run locally:

# Build image
docker build -t chicken-disease-classifier-cnn .

# Run container (expose Flask on 8080)
docker run -p 8080:8080 chicken-disease-classifier-cnn

Then open http://127.0.0.1:8080/.

---

Prediction Examples

Examples from the ingested dataset (for illustration):

• Coccidiosis: artifacts/data_ingestion/Chicken-fecal-images/Coccidiosis/cocci.0.jpg
• Healthy: artifacts/data_ingestion/Chicken-fecal-images/Healthy/healthy.0.jpg

---

Development Notes

• Package install is editable via -e . in requirements.txt (see setup.py).
• The Flask /train route runs dvc repro server-side to retrain using your current configs.
• Prediction pipeline loads the model from artifacts/training/model.h5 and expects 224×224 RGB.

---

Contributing

Issues and PRs are welcome. Please open an issue first for significant changes.

---

License

This project is licensed under MIT License - see the LICENSE file for details.