AI-powered restoration system for damaged NAVTEX (NAVigational TEleX) maritime safety messages.
Web interface showing real-time NAVTEX message restoration with per-character confidence scores and detailed restoration information.
This project provides the implementation of the NAVTEX message restoration model described in the following paper:
Cho, H.; Lee, C.; Lee, S. Comparative Performance Analysis of Software-Based Restoration Techniques for NAVTEX Message. J. Mar. Sci. Eng. 2025, 13, 1657. https://doi.org/10.3390/jmse13091657
If you use this code in your research, please cite our paper:
MDPI and ACS Style
Cho, H.; Lee, C.; Lee, S. Comparative Performance Analysis of Software-Based Restoration Techniques for NAVTEX Message. J. Mar. Sci. Eng. 2025, 13, 1657. https://doi.org/10.3390/jmse13091657
AMA Style
Cho H, Lee C, Lee S. Comparative Performance Analysis of Software-Based Restoration Techniques for NAVTEX Message. Journal of Marine Science and Engineering. 2025; 13(9):1657. https://doi.org/10.3390/jmse13091657
Chicago/Turabian Style
Cho, Hoyeon, Changui Lee, and Seojeong Lee. 2025. "Comparative Performance Analysis of Software-Based Restoration Techniques for NAVTEX Message" Journal of Marine Science and Engineering 13, no. 9: 1657. https://doi.org/10.3390/jmse13091657
APA Style
Cho, H., Lee, C., & Lee, S. (2025). Comparative Performance Analysis of Software-Based Restoration Techniques for NAVTEX Message. Journal of Marine Science and Engineering, 13(9), 1657. https://doi.org/10.3390/jmse13091657
BibTeX
@article{cho2025navtex,
title={Comparative Performance Analysis of Software-Based Restoration Techniques for NAVTEX Message},
author={Cho, Hoyeon and Lee, Changui and Lee, Seojeong},
journal={Journal of Marine Science and Engineering},
volume={13},
number={9},
pages={1657},
year={2025},
publisher={MDPI},
doi={10.3390/jmse13091657}
}- Purpose: AI-based restoration of damaged NAVTEX messages (masked with
*) - Model: ConstrainedCharBert (Transformer-based)
- Supported Characters: A-Z, 0-9, special characters (56 + PAD token)
- Average Accuracy: ~75% (Alphabet: 91%, Digits: 45%, Space: 95%)
- Real-time message restoration
- Per-character confidence scores with alternative predictions
- Adjustable confidence threshold (slider)
- Batch restoration support (up to 100 messages)
- Intuitive web interface (dark theme)
- Automatic GPU/CPU detection
- Processing time and error rate display
+-----------------+ HTTP/REST API +-----------------+
| | <---------------------> | |
| Vue.js 3 | | FastAPI |
| Frontend | JSON (Request/ | Backend |
| | Response) | |
| - RestoreForm | | - Model Loader |
| - ResultView | | - Restoration |
| | | Service |
+-----------------+ +--------+--------+
|
v
+-----------------+
| ConstrainedChar |
| BERT Model |
| (89MB, PyTorch) |
+-----------------+
NAVTEXRecovery/
├── backend/ # Python backend
│ ├── app/
│ │ ├── api/routes/ # API routers
│ │ ├── core/ # Configuration & model loader
│ │ ├── models/ # ConstrainedCharBert
│ │ ├── schemas/ # Pydantic schemas
│ │ ├── services/ # Business logic
│ │ └── main.py # FastAPI app
│ ├── weights/
│ │ └── best_model.pt # Trained model (89MB)
│ └── requirements.txt
│
├── frontend/ # Vue.js frontend
│ ├── src/
│ │ ├── components/ # Vue components
│ │ ├── composables/ # Composition API
│ │ ├── services/ # API client
│ │ ├── types/ # TypeScript types
│ │ ├── App.vue
│ │ └── main.ts
│ ├── package.json
│ └── vite.config.ts
│
├── model/
│ └── best_model.pt # Model checkpoint
│
└── README.md
cd backend
# Create Python virtual environment (Windows)
python -m venv venv
venv\Scripts\activate
# Create Python virtual environment (Linux/Mac)
python3 -m venv venv
source venv/bin/activate
# Install dependencies
pip install -r requirements.txt
# Verify model file exists
# backend/weights/best_model.pt should be present (89MB)
# If not, copy from parent directory:
# cp ../model/best_model.pt weights/
# Run server
python -m app.main
# or
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000API Documentation: http://localhost:8000/docs
cd frontend
# Install dependencies
npm install
# Set environment variables (optional)
# Create .env file
echo "VITE_API_URL=http://localhost:8000/api/v1" > .env
# Run development server
npm run devOpen http://localhost:5173 in your browser.
curl -X POST "http://localhost:8000/api/v1/restore" \
-H "Content-Type: application/json" \
-d '{
"corrupted_text": "ZCZC NA*8 171522 UT* SEP*3",
"return_confidence": true,
"min_confidence": 0.7
}'{
"corrupted_text": "ZCZC NA*8 171522 UT* SEP*3",
"restored_text": "ZCZC NA08 171522 UTC SEP23",
"restorations": [
{
"position": 7,
"original_char": "*",
"restored_char": "0",
"confidence": 0.9523,
"alternatives": [
{"char": "8", "confidence": 0.0312},
{"char": "O", "confidence": 0.0089}
]
}
],
"avg_confidence": 0.8756,
"mask_count": 3,
"computation_time_ms": 45.2
}- Python 3.10+
- PyTorch 2.0+ - Deep learning framework
- FastAPI 0.104 - Web framework
- Pydantic 2.5 - Data validation
- x-transformers - Transformer models
- Vue.js 3 - Progressive Framework
- TypeScript - Type safety
- Vite - Build tool
- Axios - HTTP client
| Category | Accuracy |
|---|---|
| Overall | ~75% |
| Alphabet (A-Z) | 91.1% |
| Digits (0-9) | 44.6% |
| Space | 95.0% |
| Special Characters | 85.0% |
- Architecture: ConstrainedCharBert
- Input: Single character tokens (vocab_size: 57)
- Output: Constrained character set (56 + PAD)
- Parameters: ~23M
- Model Size: 89MB
- Max Sequence Length: 1024
- d_model: 512
- num_heads: 5
- depth: 6 (Transformer layers)
- Rotary Position Embedding
- Flash Attention
- RMSNorm
- CPU: 2 cores
- RAM: 4GB
- Disk: 1GB
- CPU: 4+ cores
- RAM: 8GB+
- GPU: NVIDIA GPU (CUDA support) - optional
- Disk: 2GB+
After starting the server, API documentation is available at:
- Swagger UI: http://localhost:8000/docs
- ReDoc: http://localhost:8000/redoc
# Check if backend/weights/best_model.pt exists
ls -lh backend/weights/best_model.pt
# If not, copy from model/ directory
cp model/best_model.pt backend/weights/# Reinstall Python dependencies
cd backend
pip install -r requirements.txt
# Verify model file path
python -c "from pathlib import Path; print(Path('weights/best_model.pt').exists())"# Check environment variables
cat frontend/.env
# Verify API URL is correct
# VITE_API_URL=http://localhost:8000/api/v1
# Check CORS settings (backend/app/core/config.py)# Install PyTorch 2.0+ (CPU)
pip install torch>=2.0.0 --index-url https://download.pytorch.org/whl/cpu
# GPU version (CUDA 11.8)
pip install torch>=2.0.0 --index-url https://download.pytorch.org/whl/cu118This project is licensed under the PolyForm Noncommercial License 1.0.0.
Key points:
- Free for research, education, and non-commercial use
- Modifications allowed with proper attribution
- Not a copyleft license (your modifications don't have to use the same license)
- Commercial use requires separate licensing agreement
See the LICENSE file for full terms.
For questions or issues, please open an issue on GitHub.
NAVTEX Recovery - AI-powered Maritime Safety Communication Restoration System