The system, honestly described
What's production, what's experimental, and how the pieces fit. No marketing fluff — written for engineers, evaluators, and partners doing due diligence.
PIPELINE5-phase detection chain
Phase 1 · Ingest
Sentinel-1 IW GRD (CDSE), AisStream live feed, GFW historical, NISAR (when available), weather, ASF.
Phase 2 · Detection
SAR calibration (SNAP / Rust speckle filter), YOLO ship detector, CFAR thresholding.
Phase 3 · Intelligence
AIS correlation, spoofing checks, STS proximity (500m rule), geofence breaches, alerts.
Phase 4 · Report
PDF dossier per detection, vessel profile, GeoJSON export.
Phase 5 · Analytics
NISAR wake analysis, Kelvin-wake speed regression, Froude validation.
Phase D · Multi-sensor
VIIRS bioluminescence, Landsat thermal, Sentinel-2 glint, CYGNSS GNSS-R — cross-confirmation hooks.
PROCESSORSSAR calibration backends
| Backend | Runtime | Temp files | Status |
|---|---|---|---|
| SnapProcessor | ~750s/scene | 4 GB | PRODUCTION |
| RustProcessor | ~780s/scene | 0 | PRODUCTION |
| CppProcessor | 60–120s (target) | 0 | PHASE 4b · WIP |
Pixel-level accuracy (Rust vs SNAP): max diff < 0.1 dB, RMSE < 0.05 dB. YOLO detection consistency: ≥95% spatial overlap, KS-test p > 0.05.
HONESTYWhat is and isn't ready
LICENSINGCommercial considerations
YOLO (Ultralytics)
Currently AGPL v3 — isolated to detect/ships.py. ONNX export + C++ inference path planned before any closed-source deployment.
SAR data
Sentinel-1 (Copernicus, free, commercial use OK). NISAR (NASA/ISRO, open). Commercial high-res via partnership.
AIS data
AisStream (free, terms-of-service constrain redistribution). Spire / Orbcomm available for commercial-grade feeds.
Technical questions?
We answer in plain language. No NDA required for general architecture discussion.
Email us