All projects

CASE STUDY / COMPUTER VISION

TritonEye

Maritime Detection & Geospatial Evaluation

Maritime inference, tested against its limits.

Research project

01 / The problem

Satellite vessel detections must be geolocated and evaluated against explicit reference conditions before their usefulness can be assessed.

02 / My approach

The public evaluation investigates geolocation, image preprocessing, confidence thresholds, and detector weights against AIS-derived reference data.

Conceptual workflow
  1. SAR imagery
  2. Geolocation
  3. Detection
  4. AIS-based evaluation

03 / Engineering decisions

  • Inspect geolocation before interpreting detection quality.
  • Publish weak detector results and failure analysis.
  • Separate scene-specific evaluation from operational readiness.

04 / Results & current state

Research pipeline; a publicly operating service is not verified.

SAR detection investigation

Historical report: 2026-08-18. Date stated in the evaluation report

The report documents low recall for the original YOLOv8 configuration and compares an alternative detector. Read the report for scene and threshold conditions.

Original maritime evaluation report

05 / Limits of the evidence

Historical evaluation reports substantial detector limitations; engineering work does not establish operational readiness. Results are specific to the report’s scenes, vessel-size filters, thresholds, and AIS-derived reference data; not general maritime detection accuracy.

Public sources

Reviewed September 8, 2026. Links are pinned to the inspected public repository revision.

Project documentationHistorical evaluation report
Ask my twin