Tasmanian Scientists Train AI to Recognise Individual Southern Rock Lobsters

Researchers at the University of Tasmania’s Institute for Marine and Antarctic Studies have built an artificial intelligence system that can tell one southern rock lobster from another by the patterns on its shell, ABC News reported. Associate professor Quinn Fitzgibbon compared it to a fingerprint or facial recognition: a quick smartphone photo either matches a lobster already in the database or adds a new one to a collection that now holds thousands of animals from Australia and New Zealand.

The work, by Fitzgibbon, Dean Giosio and Tara Kelly, is being done with the Fiordland Lobster Company and the South Australian Lobster Company. In testing, the model picked out the same lobster three years later, after it had moulted into a fresh shell or gathered barnacles and algae, and could even recognise it once cooked. Fitzgibbon said being able to trace a lobster from the point of capture to a market or restaurant in Sydney or China would be a first.

The stakes are commercial. Tasmanian fishers land just over 1,000 tonnes of southern rock lobster a year, worth about $90 million, and the industry is recovering from trade restrictions and a glut of cheap product at home. Bicheno exporter Michael Blake, who processes up to 300,000 lobsters a year, told the ABC that once exported they are often mixed in tanks with lobsters from New Zealand, South Africa and elsewhere, so buyers cannot tell what they are getting. Physical tags are no answer, because they can be removed or forged.

What it means in practice

The project is a reminder that some of Australia’s most practical AI sits far from the data-centre and regulation debates. It relies on the kind of shared research data described in universities opening shared datasets for applied machine learning, and fits the productivity case made in business councils’ estimates of AI’s contribution to output by 2035.

Next, the team wants the system to log sex, size, colour, defects and eventually vitality for processors, which matters for live lobsters making the long trip to China. IMAS senior research fellow John Keane said near-real-time data from almost 100 commercial boats, compared with the one or two observers IMAS usually has at sea, would give more assurance that advice to government on the catch is accurate and sustainable.

What to watch: whether exporters adopt photo checks at the dock, whether consumers get a way to scan their own dinner, and whether digital records start to replace physical tags in fishery surveys.