A pioneering project led by the University of Cape Town’s Department of Biological Sciences is developing artificial intelligence (AI) tools to improve onboard monitoring of fish catches in South Africa’s trawl fishery for hake.
By combining AI with video technology, the project aims to deliver more accurate and transparent data on what’s being caught at sea.
This initiative is one of three supported by a £25 000 (approximately R610 000) grant from the Marine Stewardship Council’s Ocean Stewardship Fund. The grant was awarded to the South African Deep-Sea Trawling Industry Association for forward-looking research in sustainable fisheries management.
The AI software analyses footage of the catch as it moves over conveyor belts in onboard fish factories. The software automatically identifies and quantifies catches of hake – the target of the fishery – and species such as kingklip, monk and horse mackerel which are caught in trawl nets alongside hake. Collectively, these fishes are known as “non-target species” or “bycatch” and they are retained, processed and sold on local and international markets.
Improving knowledge about bycatch will improve the management of the trawl fishery for hake as a whole, says project lead Colin Attwood, associate professor in the Department of Biological Sciences at the University of Cape Town:
“First prize will be the improvement in the quality and the volume of the information. We want to know what is being caught. If we know what’s being caught, we can adjust the models and we can make recommendations. The more reliable the information, the better we can act on it,” he says.
Traditionally, fisheries rely on three main sources of data: fish landed and recorded at harbours, information recorded by onboard observers, and government research surveys. Observers play a key role by sampling catches, identifying and weighing species and recording discards. But observers are human – they cannot monitor around the clock and they are often the sole data collectors on board a fishing vessel, meaning their coverage is limited and subject to bias and fatigue.
To address this, the project is testing a new system: cameras installed on vessels that automatically record fish as they move over conveyor belts in the fish factory. The footage is then analysed using AI software trained to identify species based on thousands of annotated video frames. The AI uses shape, colour and size to identify species and assign confidence levels to each detection.
Cameras installed on trawlers (top) automatically record fish as they move over conveyor belts in the fish factory (bottom). The footage is analysed using AI software trained to identify species based on thousands of annotated video frames
The approach not only boosts data reliability but also frees observers to focus on tasks that require human judgment, ultimately enhancing fisheries management and sustainability.
So far, the system has been trained on around 10 common species, with more to be added. The advantage of AI is its consistency, full coverage and ability to process large volumes of data quickly and objectively. The videos are archived for future review, allowing for transparency and verification.
Phase one of the project has focused on factory trawlers that produce fillets and other hake products at sea. The team has recently rolled out cameras to “wetfish” trawlers that preserve the catch on ice for processing on land. The next step is to refine the AI’s reporting functions to produce clear summaries of catch composition.
While AI won’t replace human observers who will still perform critical tasks like checking gear compliance and observing crew behaviour, it can reduce their workload and improve overall data quality. The long-term goal is to hand over the system to the industry and government regulators with the goal of enhancing fisheries management and sustainability through the provision of more accurate, verifiable data.
“We want good data. The more reliable the information, the better the models, and the better the decisions. This is about improving the whole system,” says Colin.
Although similar projects have been trialled internationally, this is the first of its kind in the South African fishing industry. With growing interest from industry and the scientific community, the project promises to set a new standard for monitoring, accountability and sustainable fishing.
