Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Satellite Imagery and AI: A New Era in Ocean Conservation, from Research to Deployment and Impact (Version. 2.0)

Patrick Beukema, Favyen Bastani, Yawen Zheng, Piper Wolters, Henry Herzog, Joe Ferdinando

TL;DR

This work addresses the need for global, near-real-time monitoring of IUU fishing by deploying four sensor-specific computer vision pipelines (VIIRS, SAR from Sentinel-1, optical from Sentinel-2, and Landsat) within Skylight, processing vast satellite data to detect vessels with high precision. It introduces a mix of hybrid and deep learning architectures tailored to each sensor, including a three-stage VIIRS pipeline, a SAR-based Faster-RCNN with historical-context augmentation, a Swin Transformer-backed S2 detector, and a two-stage Landsat detector/classifier, all open-sourced and continuously evaluated through offline and online stages. The approach emphasizes latency, interpretability, and robust deployment practices (CI/CD, staging, and user feedback) and includes explicit strategies for reducing false positives via postprocessing and geofencing, as well as aligning detections with AIS data. The work demonstrates practical impact by providing a free, globally accessible maritime monitoring platform that supports conservation efforts while acknowledging potential misuse and leveraging open data ecosystems for scalable ocean stewardship.

Abstract

Illegal, unreported, and unregulated (IUU) fishing poses a global threat to ocean habitats. Publicly available satellite data offered by NASA, the European Space Agency (ESA), and the U.S. Geological Survey (USGS), provide an opportunity to actively monitor this activity. Effectively leveraging satellite data for maritime conservation requires highly reliable machine learning models operating globally with minimal latency. This paper introduces four specialized computer vision models designed for a variety of sensors including Sentinel-1 (synthetic aperture radar), Sentinel-2 (optical imagery), Landsat 8-9 (optical imagery), and Suomi-NPP/NOAA-20/NOAA-21 (nighttime lights). It also presents best practices for developing and deploying global-scale real-time satellite based computer vision. All of the models are open sourced under permissive licenses. These models have all been deployed in Skylight, a real-time maritime monitoring platform, which is provided at no cost to users worldwide.

Satellite Imagery and AI: A New Era in Ocean Conservation, from Research to Deployment and Impact (Version. 2.0)

TL;DR

This work addresses the need for global, near-real-time monitoring of IUU fishing by deploying four sensor-specific computer vision pipelines (VIIRS, SAR from Sentinel-1, optical from Sentinel-2, and Landsat) within Skylight, processing vast satellite data to detect vessels with high precision. It introduces a mix of hybrid and deep learning architectures tailored to each sensor, including a three-stage VIIRS pipeline, a SAR-based Faster-RCNN with historical-context augmentation, a Swin Transformer-backed S2 detector, and a two-stage Landsat detector/classifier, all open-sourced and continuously evaluated through offline and online stages. The approach emphasizes latency, interpretability, and robust deployment practices (CI/CD, staging, and user feedback) and includes explicit strategies for reducing false positives via postprocessing and geofencing, as well as aligning detections with AIS data. The work demonstrates practical impact by providing a free, globally accessible maritime monitoring platform that supports conservation efforts while acknowledging potential misuse and leveraging open data ecosystems for scalable ocean stewardship.

Abstract

Illegal, unreported, and unregulated (IUU) fishing poses a global threat to ocean habitats. Publicly available satellite data offered by NASA, the European Space Agency (ESA), and the U.S. Geological Survey (USGS), provide an opportunity to actively monitor this activity. Effectively leveraging satellite data for maritime conservation requires highly reliable machine learning models operating globally with minimal latency. This paper introduces four specialized computer vision models designed for a variety of sensors including Sentinel-1 (synthetic aperture radar), Sentinel-2 (optical imagery), Landsat 8-9 (optical imagery), and Suomi-NPP/NOAA-20/NOAA-21 (nighttime lights). It also presents best practices for developing and deploying global-scale real-time satellite based computer vision. All of the models are open sourced under permissive licenses. These models have all been deployed in Skylight, a real-time maritime monitoring platform, which is provided at no cost to users worldwide.
Paper Structure (11 sections, 3 figures, 1 table)

This paper contains 11 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Building computer vision models for maritime intelligence
  3. Vessel detection in VIIRS imagery
  4. Vessel detection in S1 imagery
  5. Vessel detection in S2 imagery
  6. Vessel detection in Landsat imagery
  7. Model evaluation, validation, and deployment
  8. Known Limitations
  9. Correlating satellite detections and GPS
  10. Best practices for shipping computer vision for maritime intelligence
  11. Equity considerations

Figures (3)

  • Figure 1: Data-flow depiction of a real-time streaming computer vision service for vessel detection in satellite imagery. An orbiting satellite images a vessel. The image is downlinked to a ground station, copied to Skylight owned servers and processed by a computer vision model netron. The resulting vessel detection is reported to our users through a GUI and available via an API.
  • Figure 2: Example satellite imagery (top row) and sample detections (bottom row) from a VIIRS image (A, E) near the Ecuadorian coast, an S1 image (B, F) from the North Sea, an S2 image (C, G) from the Maldives, and a Landsat image (D, H) from the Suez Canal. Scale bars are approximate. Confidence scores $>$ 0.95.
  • Figure 3: A. Depiction of the correlation process. We compute the haversine distance between vessels in imagery and as located by AIS, then minimize the distance over the pairs. B. Panel from the Skylight UI showing radar, optical, and night lights detections near Colón, (Panama Canal). Correlated = black, uncorrelated = red.