Coastal drones use artificial intelligence to save us from worrying about shark threats

Along the Australian coastline, we often see drones hovering overhead, shooting straight towards the sparkling turquoise water. “An internet celebrity must be filming here again,” we complained in our hearts. But if you look carefully at this aircraft, you will find a sign painted on it, which reads "Keep Clear" in bright yellow and red letters. Therefore, this should not be posted by TikTok broadcasters.

Yes, it is not a toy for Internet celebrities, but a savior for surfing enthusiasts. The drones track sharks on beaches to prevent these dangerous creatures from getting too close to swimmers. The New South Wales government is ready to invest heavily (more than $85 million) in the next few years to reduce the threat of sharks and allow humans to better coexist with these predators. Of course, traditional crane patrols and shark nets still exist, but a 2020 survey showed that the public's favorite shark management solution is still this kind of drone tracking.

This initiative is not new. The state government has been using drones to hunt for sharks since 2016, working in partnership with Surf Life Saving NSW since 2018. The drone flies at an altitude of more than 60 meters and is operated by a trained surf lifesaving pilot. The real-time video taken by passing over the ocean again and again is not to record the beautiful waves, but to search for sharks flowing under the water. Operators are of course trained to differentiate between shark morphology (great whites and tiger sharks are clearly different, for example) and other animals (seals or large fish), but if weather conditions are poor, visual judgment can be quite difficult. Once the wind blows, strong light hits a specific location, the water is too murky or dark, or you fly through a seaweed-rich area, the shot will become even more difficult to distinguish.

Photo taken on December 10, 2017, shows a shark surveillance drone flying over Bilgola Beach, north of Sydney, with a safety flotation device underneath the fuselage. High-tech shark-spotting drones patrolled dozens of Australia's beaches last summer, quickly identifying underwater predators and delivering safety equipment to swimmers and surfers faster than lifeguards.

The rate at which drone pilots made the right call was about 60 percent—significant, but not entirely reliable. That's why a team of scientists decided to explore whether AI could do a better job. Dr Cormac Purcell received funding from the NSW Department of Primary Industries to conduct the research at Macquarie University, alongside Dr Paul Butcher from Southern Cross University and Deakin University. The team set out to build the "most powerful" shark AI detector and immediately deployed it for testing in Australian waters. Yes, most AI technologies perform well in the laboratory, but there are a lot of challenges to deal with in real-world applications. That's why Purcell and Dr. Butcher are keen to test their detectors in the wild. They wrote in the report, "Previously, early performance of AI-enhanced shark identification systems has proved important, with such systems reporting detection accuracy of more than 90%. But expanding these systems to New South Wales beaches, in Implementing identification capabilities in real-world scenarios has always been challenging. By its nature, the AI-driven software in machine learning solutions must be regularly updated to continue to be effective."

Researchers work hard to track and identify sharks, delivering information Give AI software the ability to continuously "learn" to create a mobile app for surf lifesavers. The authors explain, "Using this new data set, we trained a machine learning model capable of identifying ten species of marine life, including different types of dangerous sharks, such as great whites and whale sharks. We then embedded the model into a new mobile app that tags sharks captured in drone footage and their corresponding species in real time. We worked closely with the New South Wales Government and Surf Lifesaving NSW to trial this app at five beaches in summer 2020 Application."

How does it perform specifically? "Our AI shark detector performs well. Under real-world conditions, it can identify dangerous sharks frame by frame with an 80% success rate. We also deliberately buried holes for it, by using images from different times of the year or under different weather conditions. Beach scenes are used to increase the difficulty of the test." Currently, the app still has some limitations, such as difficulty in identifying different shark species with similar outlines, and it is easy to miss smaller creatures. But the team is convinced that AI "is now mature enough to be deployed on Australian beaches to shoulder the task of detecting sharks. But unlike conventional software, this solution needs to be regularly monitored and updated to maintain highly reliable detection of dangerous sharks." Capability."

It is summer in Australia, and drones have once again gone out to scan the sea and contribute to the protection of beachfront tourists. The authors concluded, "AI can improve the monitoring reliability of these drone devices and may even develop into stable and reliable fully automated shark detection and response tools in the future."

The above is the detailed content of Coastal drones use artificial intelligence to save us from worrying about shark threats. For more information, please follow other related articles on the PHP Chinese website!