Wadden Sea Seal Monitoring

The Wadden Sea — a UNESCO World Heritage Site spanning the Dutch, German and Danish coasts — is one of Europe’s most important nurseries for grey seals and harbour seals. To track the health of those colonies, researchers at Wageningen Marine Research photograph them from light aircraft several times a year, building a record of population size, age structure and breeding success.

But turning those photos into numbers has meant hundreds of hours of painstaking work — counting and classifying every seal by hand. Together with Wageningen Marine Research and Lumax AI, we built an AI system that does it automatically: it detects, counts and classifies every seal in an aerial image, and even recognises individuals by their natural markings — freeing researchers to spend their time on the science.

1 / 3

From a survey flight to a validated count, the system runs in four steps:

Aircraft photograph the colony; a model finds every seal, classifies each one five ways, and the counts are reviewed and exported in a web app.

Why seals matter

Seals sit near the top of the Wadden Sea food web, which makes them a sensitive barometer for the health of the whole ecosystem.

The monitoring challenge

Doing this well, by hand, is genuinely hard. Tap each challenge to see why.

Two systems working together

Two AI systems work side by side, across both grey seals (Halichoerus grypus) and harbour seals (Phoca vitulina): one counts and classifies every seal in a photo, the other recognises individuals over time. Both feed a single web app where researchers check the AI’s work and export validated data.

Counting and classifying every seal

The first system starts with a detector that finds every seal in an aerial photo — even animals that are partly submerged, overlapping, or caught in awkward light. It then classifies each one five different ways:

Every prediction lands in a web app that keeps researchers firmly in control: detected seals are drawn with colour-coded labels for all five attributes, uncertain cases are flagged for a closer look, any label can be fixed (or a missed seal added) in one click, and the validated counts export straight to a spreadsheet.

1 / 3

Recognising individual seals

Tracking individual animals usually means tagging them — capture, handling, stress for the seal and logistics for the team. Our second system avoids all of that. Seals carry stable, distinctive markings — whisker spots, coat patterns, facial features — that stay with them for life, so the AI can recognise an individual from a photo alone, the way facial recognition works for people.

The model turns each seal’s face into a compact “fingerprint”, then compares it against a database of known individuals to find the best match.

When a new photo comes in, the model condenses the seal’s face into that fingerprint and compares it with every seal already on file, returning the closest matches for a researcher to confirm — or registering a brand-new individual. The matches below were all made this way, from natural markings alone:

1 / 3

Recognising individuals over time turns a population count into a living record:

The impact

The system has reshaped the work at Wageningen Marine Research. Analysis that once took hundreds of hours now takes a fraction of the time, and the counts are more consistent from one survey year to the next. Researchers can focus on biology and conservation strategy instead of data entry — and the re-identification system opens an entirely new, individual-level view of seal life that simply wasn’t possible before. The same framework can extend to other colonies, regions, or marine-mammal species.