Just in time for summer, surfers and swimmers flocking to Australia’s beaches have a new ally keeping them safe from shark encounters: drones.
A team from University of Technology Sydney and industry partner the Ripper Group are developing drones that use artificial intelligence to spot sharks in coastal waters.
Drones have been used in the past to patrol Australian beaches, but this is the first time it’s being coupled with AI to detect sharks in real time.
Although shark attacks are relatively rare, almost 30 per cent of attacks worldwide happen along Australia’s 25,000 km of coastline. This makes Australia the perfect testing ground for the technology, said Professor Michael Blumenstein, associate dean (research strategy and management) and head of the School of Software at University of Technology Sydney.
The drones – called Little Ripper Lifesavers – use a deep learning algorithm and a region-based convolutional neural network (RCNN) for object detection and recognition. To train the software, researchers fed it a steady diet of aerial footage captured from publicly available sources and by the drone itself.
“We have to get a lot of data to teach the network so it can be trained and learn to identify the different types of samples. Because of its sophistication, with a reasonable number of samples, it can learn to accurately tell the difference between a dolphin and a shark, or a shark and a whale,” Blumenstein said.
The software thrives on large amounts of data input. After more than a year of training, it now boasts an accuracy rate of 90 per cent, according to Blumenstein. This is a massive lead over human shark spotters, as a person’s ability to distinguish shapes can be compounded by environmental factors like glimmers on water or the angle of the footage.
“Humans can be significantly less accurate because humans get tired, they make errors in judgement and they don’t deal well with the subtleties of video data,” Blumenstein said.
After completing a number of trials, the drones were deployed at the official ‘flag raising’ ceremony for the start of surf lifesaving season at Sydney’s Maroubra Beach. Since then the drones have been assisting lifeguards at beaches in the Port Macquarie-Hastings region. They clock in at 8:30am and patrol the beach at 30-minute intervals, looking for those telltale shark silhouettes.
The plan is to steadily increase the drones’ workload until it’s capable of full-time deployment. However, Blumenstein insisted the project’s objective is not to replace humans altogether, but rather to work alongside them.
“There are some things humans do better; there’s instinct, experience and other things we can’t replicate in machines. We always have the objective of assisting and supporting the people who save lives on beaches,” he said.
Aside from shark spotting, Blumenstein said the technology can also be used for environmental and social research, such as tracking marine migrations or preventing drownings.
“We’re very open to see what other problems can be solved with a similar solution, and how we can generate benefits from a societal and environmental point of view,” he said.
“That’s nice for me as a researcher – to work on something that can have that kind of impact.”