🌿 AgriVision — Crop Disease Detection & Advisory System

🚀 Built with PyTorch, FastAPI, and Streamlit for real-time crop disease detection.

An end-to-end AI-powered decision support system that detects crop diseases from leaf images and provides actionable treatment recommendations.

---

🎯 Problem

Crop diseases can significantly reduce yield, and early detection is crucial. Manual diagnosis is time-consuming and requires expertise.

AgriVision provides an automated solution using deep learning for quick and accurate disease detection.

---

🚀 Features

• Multi-crop classification (Tomato, Potato, Corn — 9 classes)
• Top-3 predictions with confidence scores
• Interactive UI with image preview and confidence visualization
• Rule-based treatment recommendations
• FastAPI backend for real-time inference
• Model evaluation (confusion matrix, precision, recall, F1-score)

---

🧠 Model

• Architecture: Custom CNN (5 convolutional blocks)
• Input Size: 224 × 224 RGB
• Dataset: PlantVillage (curated subset)
• Validation Accuracy: ~98%

---

📂 Project Structure

agrivision/
├── backend/        # FastAPI server
├── frontend/       # Streamlit UI
├── models/         # CNN model, training, evaluation
├── inference/      # preprocessing + predictor
├── data/           # dataset
├── assets/         # screenshots
├── requirements.txt
└── README.md

---

📈 Key Results

• Achieved ~98% validation accuracy
• Strong performance across all 9 classes
• Reliable top-3 predictions for ambiguous cases
• Deployed as a real-time system with API + UI integration

---

🧩 Tech Stack

Layer	Technology
ML	PyTorch
Backend	FastAPI, Uvicorn
Frontend	Streamlit
Utilities	NumPy, scikit-learn, PIL

---

🏗️ Architecture

Streamlit UI  →  FastAPI (/predict)  →  PyTorch Model
                                       ├─ Softmax
                                       └─ Top-3 Predictions

---

📸 Screenshots

🌿 Streamlit UI

Upload & Preview	Predictions

---

⚙️ FastAPI (Swagger)

Endpoint Testing	API Response

---

▶️ How to Run

1. Clone repository

git clone https://github.com/snehapriy958/agrivision.git
cd agrivision

2. Setup environment

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt

3. Start backend

uvicorn backend.main:app --reload

4. Start frontend

streamlit run frontend/app.py

Access

• UI → http://localhost:8501
• API → http://127.0.0.1:8000/docs

---

📊 Example Output

{
  "predictions": [
    {"label": "Tomato___Early_blight", "confidence": 0.9989},
    {"label": "Tomato___Late_blight", "confidence": 0.001},
    {"label": "Tomato___healthy", "confidence": 0.0001}
  ]
}

---

🧪 Evaluation

• Confusion matrix and classification report
• Strong performance across all classes
• Minor confusion between visually similar diseases

---

💡 Future Improvements

• Transfer learning (EfficientNet)
• Grad-CAM for explainability
• Multi-language UI
• Batch prediction support

---

👤 Author

Sneha — AIML Engineer

---

⭐ If you like this project

Give it a star ⭐ and share your feedback!