Video surveys and artificial intelligence (AI) are transforming scallop monitoring, delivering faster, more accurate data to support sustainable fisheries. 

An FRDC project is helping to change how Commercial Scallop (Pecten fumatus) populations are monitored by swapping dredges for video monitoring, using artificial intelligence (AI) and machine learning to boost results.

Project 2021-010, led by Ryan Day from the Institute for Marine and Antarctic Studies (IMAS), has developed a suite of low-cost video systems and machine learning tools that are set to make scallop surveys faster, cheaper and more sustainable.

A clearer view beneath the surface

Traditional scallop surveys rely on dredging to collect data on stock abundance and size structure. While effective, dredging is time-intensive, selective for larger scallops and can impact sensitive seabed habitats.

Ryan says that video technology offers a great alternative. “Video allows us to assess the whole population, from the juveniles to adults, without disturbing the seabed. It’s non-invasive, scalable and cheaper to run than dredge-based surveys.”

Using compact GoPro cameras mounted on custom-built frames, the team trialled two survey systems: a towed video rig for calm, inshore conditions, and a drop camera setup for deeper, offshore areas where towing isn’t possible. Both systems record high-resolution footage of scallop beds, giving researchers a detailed view of density, distribution and size.

“These setups are deployed from small vessels and upgraded easily as technology improves,” Ryan explains. “It means we can keep pushing the science forward without massive cost barriers.” 

Recording the footage is only half the battle. Analysing hours of video, frame by frame, has traditionally taken weeks, time that fishery scientists can’t always spare.

That’s where AI comes in. The team partnered with CSIRO to develop a machine learning model that automatically identifies scallops, distinguishes whether they are alive or dead and even measures their size.

“The model performs as well as a human analyst, with over 95 per cent accuracy, but it can process results in hours rather than weeks,” Ryan says.

The AI proved particularly strong at recognising whether scallops were alive or dead, a critical factor in assessing the health of a bed. Measuring size, however, remains a little trickier. With more training data from future surveys, the team expects the system to become even more precise.

Interestingly, the researchers found that even small upgrades in camera technology could affect AI accuracy. “Footage from different GoPro models varied enough that we had to retrain the system. It was a great reminder of how important consistent, high-quality data is when using AI,” Ryan adds. 

From Tasmanian waters to the world

The technology was tested in both the Tasmanian Scallop Fishery (TSF) and the Bass Strait Central Zone Scallop Fishery (BSCZSF), including a survey inside the Boags Marine Park, where dredging is prohibited.

“Being able to collect data in protected areas for the first time is a big win,” Ryan says. “We can now survey places that were completely off-limits to traditional methods.”

The results are already sparking international interest, with fisheries in places like New Caledonia reaching out to learn more. The low-cost, open-source design of the system makes it easy for other small fisheries to adopt and adapt the tools.

“We’re already seeing the potential to apply these same systems in other habitats,” Ryan says. “If we can adapt the AI models and keep refining the hardware, it could become a universal toolkit for underwater monitoring, one that’s affordable enough for smaller fisheries and research programs to use worldwide.” 

The road ahead

With proof of concept achieved, the next step is refining how video survey data can integrate with existing management rules, such as discard rate thresholds that were designed around dredge data.

“Video gives us a much clearer picture, including all the small scallops that dredges miss,” Ryan says. “The challenge now is adjusting management frameworks to make full use of the data.”

The team also hopes to see the technology rolled out more widely, supporting small-scale fisheries where funding and capacity for traditional surveys are limited.

“This project shows that innovation doesn’t have to be expensive,” says FRDC Research Portfolio Manager Adrianne Laird. “By combining simple tools with smart analytics, we can help fisheries make better, faster decisions.”