🌿 WeedScan — AI-Powered Weed Detection System

A real-time weed and crop detection system using YOLOv8 and edge computing.
The AI model runs entirely on your PC — no cloud, no internet required during scanning.

---

What it does

Point your phone camera at a field and WeedScan instantly detects and classifies plants as weeds (orange boxes) or crops (green boxes) in real time. All inference happens on your PC over your local WiFi network.

---

Screenshots

Live Detection	Analytics	Scan Report
Camera feed with bounding boxes	Charts, stats, insights	Full report with PDF export

---

System Requirements

PC (runs the AI backend)

Component	Minimum	Recommended
OS	Windows 10	Windows 11
CPU	Intel Core i3 / AMD Ryzen 3	Intel Core i5 or better
RAM	4 GB	8 GB
Storage	2 GB free	5 GB free
Python	3.9 – 3.14	3.11
Node.js	18+	20+

Mobile Phone (accesses the app)

Requirement	Details
OS	Android 8.0+ or iOS 14+
Browser	Chrome 90+ (Android) / Safari 14+ (iOS)
Camera	Any rear camera
Network	Same WiFi as the PC
RAM	2 GB minimum
Storage	No app install needed — runs in browser

Note: The phone only needs a browser. No app installation required.
Camera access requires enabling the insecure origin flag in Chrome (see Setup below).

---

Project Structure

Weed detection system/
├── start.bat                 ← Double-click to start everything
├── merge_and_train.py        ← Retrain model with new data
│
├── backend/
│   ├── main.py               ← FastAPI inference server
│   ├── best.onnx             ← Trained YOLOv8s model
│   └── scans.db              ← Local scan history (SQLite)
│
├── dataset/                  ← Training dataset (crop & weed images)
│   ├── train/
│   └── valid/
│
└── weedscan-web/             ← React frontend
    ├── src/
    │   ├── App.jsx
    │   ├── api.js
    │   └── screens/
    │       ├── LiveDetection.jsx
    │       ├── Analytics.jsx
    │       ├── CaptureReport.jsx
    │       ├── History.jsx
    │       └── ScanReports.jsx
    └── package.json

---

Setup & Running

1. Install dependencies

Python packages:

pip install fastapi uvicorn onnxruntime pillow numpy python-multipart

Node.js packages:

cd weedscan-web
npm install

2. Start the app

Simply double-click start.bat — it starts both the backend and frontend automatically.

Or start manually:

# Terminal 1 — Backend
cd backend
python -m uvicorn main:app --host 0.0.0.0 --port 8000 --reload

# Terminal 2 — Frontend
cd weedscan-web
npm run dev

3. Open on PC

http://localhost:5173

4. Open on Phone

Connect your phone to the same WiFi as your PC, then open:

http://192.168.1.8:5173

Replace 192.168.1.8 with your PC's actual WiFi IP.
Find it by running ipconfig in PowerShell and looking for the WiFi IPv4 address.

---

Phone Camera Setup (Chrome on Android)

Chrome blocks camera access on plain http:// addresses for security.
Do this once to enable it:

1. Open Chrome on your phone
2. Go to: chrome://flags/#unsafely-treat-insecure-origin-as-secure
3. In the text box, enter: http://192.168.1.8:5173
4. Set the dropdown to Enabled
5. Tap Relaunch
6. Open http://192.168.1.8:5173 and tap Allow when asked for camera

---

Features

📷 Live Detection

• Real-time camera feed with bounding boxes
• Orange boxes = weeds, Green boxes = crops
• Live FPS counter and inference time
• Animated scan line and corner bracket overlay
• Recent scans accessible without leaving the camera

📊 Analytics Dashboard

• Time filters: Today / 7 Days / 30 Days
• Stat cards: total weeds, avg confidence, inference speed, total scans
• Bar chart: weeds vs crops per day
• Auto-generated insights based on scan history

📸 Capture & Report

• Freeze frame with bounding boxes drawn on snapshot
• Full metrics: weed count, crop count, density %, confidence %, inference time
• Export to PDF (opens browser print dialog → Save as PDF)
• Share report as text
• Save annotated image to device

📋 Scan Reports

• Full history of all past scans
• Search and filter by weed/crop presence
• Expandable cards with full snapshot and stats
• Delete individual scans with confirmation

---

Model Details

Property	Value
Architecture	YOLOv8s
Format	ONNX
Input size	640 × 640
Classes	crop, weed
Training images	~1,300 labeled images
Confidence threshold	0.25
Inference time	~120–160 ms/frame on CPU

---

Retraining the Model

To improve accuracy with more data:

1. Add new labeled datasets (YOLOv8 format) to the dataset/ folder
2. Run the training script:

   python merge_and_train.py

3. The script automatically merges datasets, trains, exports to ONNX, and copies the model to backend/best.onnx
4. Restart the backend — it picks up the new model automatically

For faster training, use Google Colab with a free T4 GPU (~15 min vs ~8 hours on CPU).

---

Tech Stack

Layer	Technology
AI Model	YOLOv8s (Ultralytics)
Inference	ONNX Runtime
Backend	FastAPI + Uvicorn
Frontend	React + Vite
Charts	Recharts
Database	SQLite (via Python)
Networking	Local WiFi (no internet needed)

---

Troubleshooting

Camera not working on phone
→ Follow the Chrome flags setup above. Camera requires the insecure origin flag on HTTP.

"Site can't be reached" on phone
→ Make sure phone and PC are on the same WiFi network. Check your PC's IP with ipconfig.

No detections showing
→ Ensure the backend is running (start.bat). Point camera at real plants in good lighting.

High inference time (>300ms)
→ Normal on low-end CPUs. Close other applications to free up CPU resources.

PDF not downloading
→ The PDF button opens a print dialog. Select "Save as PDF" as the printer destination.

---

License

For educational and research use.