Ask Global Fishing Watch
Small Vessel Detection
Yuexin Li, Machine Learning Engineer, Remote Sensing
How Does AI Detect Small Vessels in Satellite Imagery?
There are more than 2 million small-scale fishing vessels operating around the world and most rarely broadcast their position on the automatic identification system, creating one of the most persistent blind spots in ocean monitoring.
Global Fishing Watch is closing this gap using Planet high-resolution imagery, which is capable of detecting small vessels under 20 meters, and rethinking our machine learning models to interpret the data.
Global Fishing Watch plans to share a small-vessel detections dataset with research partners in a future release.
How It Works
Stage 1 – Raw Detection: At Planet’s three-meter pixel images, vessels under 20 meters, such as small-scale fishing pirogues in West Africa (top image) or recreational boats in the Red Sea (bottom image), occupy only a few pixels per image. In capturing these challenging small craft, our detector inevitably picks up environmental noise like sun glint, whitecaps, or vessel wakes. The result is a raw pool of potential vessel detections, each paired with an estimated score, length, speed and heading.
Stage 2 – Noise Classification: Images of each possible vessel detection in that noisy pool are classified, incorporating different image embeddings — numerical vectors produced by an AI model that translate visual patterns like shape, size and texture into computer-readable data — and contextual features describing the environment where the detection is located, such as distance from shore, wind speed, etc.
Stage 3 – Vessel Classification: This step distinguishes real vessels from other objects like waves or rocks. The resulting imagery provides enough visual and contextual information to enable downstream models to separate fishing from non-fishing vessels, and even identify specific activities like trawling or dredging.
Stage 4 – Applied Analysis: The data is now ready for analysis. For instance, mapping previously unseen small-vessel activity along African coastlines reveals vessel traffic from boats under 20 meters related to fishing and tourism around Kenya’s Kisite Mpunguti Marine National Park. This data allows scientists and policymakers to assess the pressure that critical ecosystems face from both fishing and recreational activities.
Global Fishing Watch uses high-resolution satellite imagery and AI to map the presence of small vessels, shining a light on one of the ocean’s biggest monitoring blind spots.
©Global Fishing Watch, 2026