🌾 Silo — AI-Powered Crop Health Platform

Silo is a microservices-based platform that uses artificial intelligence to help farmers detect wheat diseases and assess crop health risk in real time.

---

🖥️ Screenshots

Dashboard — Command Centre

Real-time overview of all monitored grain silos with risk levels and sensor readings.

---

Silo Detail View — Sensor History

Per-silo view showing 24-hour temperature and humidity trends, with live sensor readings and recent alerts.

---

AI Vision Scanner — Disease Detection

Upload a grain photo directly from the silo detail page. The AI Vision service analyzes it and returns the detected disease with a confidence score.

Example result: Wheat Rust Detected — 88.5% confidence

---

📦 Project Structure

Silo/
├── ai-vision/                  # Wheat disease detection service (Keras + FastAPI)
│   ├── app/
│   │   ├── main.py             # FastAPI application
│   │   └── final_model.keras   # Trained Keras model (not tracked in git)
│   ├── train.py                # YOLOv8 training script
│   ├── requirements.txt        # Python dependencies
│   └── Dockerfile              # Container definition
│
├── ai-predictive/              # Crop health risk prediction service (XGBoost + FastAPI)
│   ├── app/
│   │   ├── main.py             # FastAPI application
│   │   └── model/
│   │       ├── predict.py      # Inference logic
│   │       └── crop_health_model_lr.pkl  # Trained model (not tracked in git)
│   ├── PREPROCESSCING/
│   │   ├── preprocess.ipynb    # Data preprocessing notebook
│   │   ├── logical_agriculture_data.csv
│   │   └── processed_agriculture_data.csv
│   ├── requirements.txt
│   └── Dockerfile
│
├── backend/                    # Main backend service
├── frontend/                   # Frontend application
├── scripts/                    # Utility scripts
└── docker-compose.yml          # Orchestrates all microservices

---

🤖 Services Overview

1. AI Vision Service (ai-vision)

Detects wheat diseases from uploaded images using a Keras deep learning model.

Property	Details
Framework	TensorFlow / Keras
Input	Wheat image (JPG/PNG)
Output	Disease label + confidence score
Port	8001
Classes	rust, blast, mildew, healthy

2. AI Predictive Service (ai-predictive)

Predicts crop health risk level based on environmental and soil sensor data.

Property	Details
Framework	Scikit-learn (Logistic Regression Pipeline)
Input	Sensor readings (JSON)
Output	Risk score + risk level
Port	8002
Classes	low, medium, high

---

🚀 Getting Started

Prerequisites

• Python 3.11+
• Docker Desktop
• NVIDIA GPU (recommended for training)
• CUDA 12.4+ (for GPU support)

1. Clone the Repository

git clone https://github.com/Ahmed-fall/Silo.git
cd Silo

2. Add Model Files

Since model files are not tracked in Git, place them manually:

ai-vision/app/final_model.keras
ai-predictive/app/model/crop_health_model_lr.pkl

3. Install Dependencies (without Docker)

AI Vision:

cd ai-vision
pip install -r requirements.txt

AI Predictive:

cd ai-predictive
pip install -r requirements.txt

---

▶️ Running the Services

Option A — Without Docker (Development)

AI Vision Service:

cd ai-vision
python -m uvicorn app.main:app --reload --port 8001

AI Predictive Service:

cd ai-predictive
python -m uvicorn app.main:app --reload --port 8002

Option B — With Docker

Build and run AI Vision:

cd ai-vision
docker build -t ai-vision .
docker run -p 8001:8001 ai-vision

Build and run AI Predictive:

cd ai-predictive
docker build -t ai-predictive .
docker run -p 8002:8002 ai-predictive

Run all services together:

docker-compose up --build

---

📡 API Reference

AI Vision Service — http://localhost:8001

GET /health

Check if the service is running.

Response:

{
  "status": "ok",
  "service": "ai-vision"
}

POST /analyze

Upload a wheat image to detect disease.

Request: multipart/form-data

Field	Type	Description
file	image	Wheat image (JPG or PNG)

Response (success):

{
  "label": "rust",
  "confidence": 0.9982,
  "status": "ok"
}

Response (model unavailable):

{
  "status": "unavailable"
}

Example using curl:

curl -X POST http://localhost:8001/analyze \
  -F "file=@wheat_image.jpg"

---

AI Predictive Service — http://localhost:8002

GET /health

Check if the service is running.

Response:

{
  "status": "ok",
  "service": "ai-predictive"
}

POST /predict

Send sensor readings to get a crop health risk assessment.

Request: application/json

{
  "temperature": 38.0,
  "humidity": 75.0,
  "soil_moisture": 25.0,
  "ndvi": 0.3,
  "pest_damage": 50.0,
  "crop_stress_indicator": 60.0,
  "soil_ph": 6.5,
  "organic_matter": 2.0
}

Field	Type	Description
temperature	float	Air temperature (°C)
humidity	float	Relative humidity (%)
soil_moisture	float	Soil moisture level (%)
ndvi	float	Vegetation index (0.0 – 1.0)
pest_damage	float	Pest damage level (0–100)
crop_stress_indicator	float	Stress indicator (0–100)
soil_ph	float	Soil pH level (0–14)
organic_matter	float	Organic matter content (%)

Response (success):

{
  "risk_score": 85.0,
  "risk_level": "high",
  "status": "ok"
}

Response (model unavailable):

{
  "status": "unavailable"
}

Example using curl:

curl -X POST http://localhost:8002/predict \
  -H "Content-Type: application/json" \
  -d '{
    "temperature": 38,
    "humidity": 75,
    "soil_moisture": 25,
    "ndvi": 0.3,
    "pest_damage": 50,
    "crop_stress_indicator": 60,
    "soil_ph": 6.5,
    "organic_matter": 2.0
  }'

---

🧪 Testing the APIs (Swagger UI)

Both services come with built-in interactive API documentation:

Service	Swagger URL
AI Vision	http://localhost:8001/docs
AI Predictive	http://localhost:8002/docs

Open in your browser, click "Try it out" on any endpoint, and test directly.

---

🏋️ Training the Model (Optional)

To fine-tune the YOLOv8 model on the wheat disease dataset:

cd ai-vision
python train.py

Requirements:

• Kaggle dataset downloaded to ai-vision/data/
• NVIDIA GPU with CUDA support
• ~1-2 hours training time

---

🔒 Notes

• Model files (*.keras, *.pkl, *.pt) are not tracked in Git due to size
• The .gitignore excludes data/, *.keras, *.pt, *.h5, __pycache__/
• Always activate your virtual environment before running locally

---

👥 Team

Service	Responsibility
ai-vision	Wheat disease detection
ai-predictive	Crop health risk prediction
backend	Main API and business logic
frontend	User interface