SKANN-SSL V2.1.0 Demo

Underwater Acoustic Vessel Classification System

Self-supervised learning approach for classifying vessel types from hydrophone audio signatures.

© 2026 Oravont Systems LLP. All rights reserved.

---

Features

• 4-Class Vessel Classification: Cargo Ship, Tanker, Fishing Vessel, Small Craft
• Real-time Audio Playback: Listen to acoustic signatures while classifying
• Radar Plot Visualization: Interactive probability display
• Physics-Aware Architecture: Selective Kernel convolutions tuned for underwater acoustics (15–500+ Hz)

---

Installation

# 1. Install dependencies
pip install -r requirements.txt

# 2. Run the demo
python skann_ssl_demo.py

Requirements

• Python 3.10+
• Windows / macOS / Linux
• ~500 MB disk space (model + sample data)

---

Folder Structure

SKANN-SSL-Demo/
├── skann_ssl_demo.py          # Main demo application
├── requirements.txt           # Python dependencies
├── README.md                  # This file
├── model/
│   ├── SKANN_SSL_Production_Bundle.joblib    # Trained encoder
│   └── vessel_territories.joblib              # Classification centroids
└── data/
    ├── manifest.csv           # Clip metadata
    └── tensors/               # Audio tensors (16 kHz, 1 second)
        ├── tensor_000000.npy
        ├── tensor_000001.npy
        └── ...

---

Usage

1. Launch: Run python skann_ssl_demo.py
2. Select Clip: Use dropdown or click "🎲 Random"
3. Classify: Click "🎯 CLASSIFY"
4. Listen: Audio plays in loop (click "🔇 Stop" to mute)
5. Review: Radar plot shows class probabilities

---

Model Performance

Metric	Value
Silhouette Score	0.8299
Embedding Dimension	128
Vessel Classes	4
Sample Rate	16 kHz
Clip Duration	1 second

Frequency Coverage (SK Kernels)

Kernel Size	Frequency	Acoustic Phenomenon
1023	15+ Hz	Shaft rate
511	31+ Hz	Generator (25 Hz)
255	62+ Hz	Generator (50 Hz)
127	125+ Hz	Blade pass
63	250+ Hz	Hull resonance
31	500+ Hz	Cavitation

---

Troubleshooting

"sounddevice not installed"

pip install sounddevice

No audio output

• Check system audio settings
• Demo works without audio (visual classification still functional)

CUDA errors on CPU machine

• The demo automatically handles CPU-only environments

---

Contact

For licensing, integration, or technical support:

Oravont Systems LLP

---

SKANN-SSL: Selective Kernel Audio Neural Networks with Self-Supervised Learning