---

🌾 DeepCrop AI 2.0 – Sugarcane Pest Detection System

DeepCrop AI 2.0 is the official Round 2 submission for AgriThon 2.0 by Team DeepCrop. It’s a full-stack, multimodal AI pipeline for detecting two major sugarcane pests — Dead Heart and Tiller — using a fusion of YOLOv8 image analysis and TabNet questionnaire classification.

This project was developed for the hackathon conducted by the School of Computer Science and Information Systems, VIT Vellore, sponsored by the Department of Biotechnology, Govt. of India.

---

📸 Screenshots & Demos

System Architecture

Web Application – Dashboard View

Prediction Flow GIF

---

🏛️ System Architecture & Pipeline

Our solution builds on the Round 1 foundation, but extends it into a production-ready web platform.

1. Data Preparation & Annotation

   • Dead Heart: Segmentation masks (YOLOv8-seg)
   • Tiller: Bounding boxes (YOLOv8)
   • Annotation handled in Roboflow + CVAT (offline).

2. Training Phase

   • Dead Heart → YOLOv8 Segmentation Model
   • Tiller → YOLOv8 Detection Model
   • Questionnaire Models → Two separate TabNet classifiers trained on curated CSV symptom datasets (>500 samples each).

3. Fusion Logic

   • Weighted combination of image score (0.6) and questionnaire score (0.4).

4. Web Application

   • Frontend: React + Vite for responsive, multilingual UI.
   • Backend: FastAPI serving YOLOv8 and TabNet inference APIs, weather API integration.

5. Deployment

   • Dockerized backend and frontend, served via Nginx for production readiness.

---

✨ Key Features

• 🧠 Multimodal Fusion: YOLOv8 + TabNet for accurate, explainable results.
• 🌿 Pest Segmentation & Detection: Precise overlays for Dead Heart, bounding boxes for Tiller.
• 🌤 Weather Risk Engine: Location-based pest risk assessment.
• 🌍 Multilingual UI: English, Hindi, Tamil, Telugu.
• ⚡ Real-time Predictions: API responses under 500ms on standard hardware.
• 📱 Field-friendly: Lightweight APIs and offline-friendly questionnaire flow.

---

🛠️ Tech Stack

Component	Technology / Library
Backend	FastAPI, PyTorch, YOLOv8, TabNet, OpenCV
Frontend	React 18, Vite, CSS3
Image Augmentation	Albumentations, Roboflow
Model Serving	Uvicorn + ASGI
Weather Integration	OpenWeather API
Containerization	Docker, Docker Compose, Nginx
Data Processing	Pandas, NumPy, Scikit-learn

---

📊 Model Performance Summary

Model	Task	Dataset Size	Metric	Inference Time
YOLOv8-seg (Dead Heart)	Segmentation	3,512 images	mAP@0.5: 89.3%	~240ms/image
YOLOv8 (Tiller)	Object Detection	3,512 images	mAP@0.5: 88.7%	~230ms/image
TabNet (Dead Heart)	Questionnaire Class.	500+ samples	Accuracy: 92.4%	~30ms/sample
TabNet (Tiller)	Questionnaire Class.	500+ samples	Accuracy: 92.4%	~30ms/sample
Fusion Output	Weighted 0.6/0.4	Combined	Final Accuracy: 94%	~300ms total

---

💻 How to Run

Prerequisites

• Python ≥ 3.9
• Node.js ≥ 16
• Git
• Docker (optional for containerized deployment)

Backend Setup

cd backend
python -m venv venv
venv\Scripts\activate   # Windows
source venv/bin/activate # macOS/Linux
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

Frontend Setup

cd frontend
npm install
npm run dev

Access:

• Frontend: http://localhost:5173
• Backend Docs: http://localhost:8000/docs

---

🐳 Docker Deployment

docker-compose up --build

---

📁 Project Structure

DeepCrop-AI-2.0/
├── backend/         # FastAPI backend
│   ├── app/
│   ├── models/
│   ├── requirements.txt
│   └── Dockerfile
├── frontend/        # React frontend
│   ├── src/
│   ├── package.json
│   └── Dockerfile
├── docker-compose.yml
└── README.md

---

👥 Team DeepCrop

Name	Email
Hariharan S	hariharan.s2022d@vitstudent.ac.in
Naresh R	naresh.r2022a@vitstudent.ac.in
Mohammed Arfath	mohammedarfath.r2022@vitstudent.ac.in
Mohammad Yusuf K A	mohammadyusuf.ka2022@vitstudent.ac.in

---

📜 License

Licensed under the MIT License – see the LICENSE file.

---

📞 Support

Open an issue on GitHub.

---