# NAVO

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Sanjeev Kumar](https://cerebralvalley.ai/u/sanjeevafk)
- **GitHub:** https://github.com/sanjeevafk/navo
- **Demo video:** https://youtu.be/4rWTBGE4WWk
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/296

NAVO (Naval Acoustic Vision & Observation) is an autonomous subsea acoustic intelligence and marine conservation platform that transforms raw side-scan sonar into pixel-exact target segmentations, ecological risk assessments, and actionable ROV salvage operations.

THE PROBLEM: 
More than 640,000 tons of ghost fishing gear, abandoned shipwrecks, and industrial debris litter the ocean floor, damaging benthic habitats and continuing to trap marine life for decades, directly implicating UN Sustainable Development Goal 14 (Life Below Water). Because sunlight does not penetrate the deep sea, autonomous underwater vehicles (AUVs) and survey vessels must rely on side-scan sonar acoustics for subsea observation. Existing detection models, however, typically produce coarse bounding boxes that combine target highlights with surrounding acoustic shadows and seabed sediment clutter. This can inflate target-area estimates by more than 36% while providing salvage teams with little insight into ecological urgency.

THE SOLUTION: 
NAVO bridges raw sonar physics with hosted Meta foundation models through an end-to-end autonomous pipeline:

1. Native Acoustic Ingestion
NAVO ingests raw binary XTF (eXtended Triton Format) sonar streams, GeoTIFF imagery, and high-frequency side-scan sweeps. The platform computes acoustic backscatter density and slant-range geometry to preserve the physical information required for downstream analysis.

2. Rapid Detection
A lightweight, edge-deployable YOLOv8-ESI detector screens survey transects using tiled inference to identify candidate anomalies in real time.

3. Meta SAM 3.1 Acoustic Segmentation
Candidate bounding boxes are passed as prompts to hosted Meta SAM 3.1 via meta-sam-parser. The model provides zero-shot acoustic generalization, isolating target highlight and shadow contours while removing 36.2% of seabed clutter. On ground-truth polygon sonar benchmarks, this increases segmentation accuracy from 0.638 to 0.840 mIoU, a 20.2% improvement.

4. Meta Muse Ecological Triage
Segmented crops, acoustic material profiles, and shadow-derived 3D heights are passed to Meta Muse (muse-spark-1.3). Muse evaluates characteristics such as synthetic netting versus metallic structures, assigns a benthic threat score from 0–100, and generates tailored low-disturbance ROV retrieval protocols.

5. Operational Command & Control
NAVO provides an interactive mission dashboard combining real-time 15 Hz acoustic waterfall feeds, CRS-aware cartographic mapping, target segmentation and ecological-risk overlays, and one-click field-deployment exports including formal PDF mission briefs, KML dive plans, and NMEA 0183 navigation waypoints.

The result is a closed-loop system that moves from raw acoustic detection → precise segmentation → ecological assessment → operational recovery, turning sonar data from a survey artifact into an actionable marine-conservation intelligence layer.

## More from Meta Global AI Developer Hackathon

- [cryptobandit](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/293)
- [Kokobubu](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/294)
- [Rafael Vanca](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/295)
- [NEURON TS LABS](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/297)
- [SHAMANTH N P : Sketch World](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/298)
- [MuseOps](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/299)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/296. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt