Avoiding Jaws: AI in Shark Detection

In case you wanted to hear even more about sharks than you got to hear in my presentation last week, I am going to use this blog post to dive a little deeper (no pun intended) into the use of technology, specifically Artificial Intelligence, used in shark detection. 

To start off, I’ll give a brief overview of the shark problem as it stands on the Cape today. Back in the early to mid 90’s, Massachusetts used to pay a bounty between $1 and $5 for every seal nose that a hunter turned in in order to keep the seal population under control. However, after the Marine Mammal Protection Act of 1972, the systematic killing of seals become a federal crime. In recent years, the seal population on Cape Cod has exploded, and unfortunately for the cape, the predators that followed them are not so cute. Cape Cod is now one of the biggest epicenters of great white shark activity in the world, with the number of shark spottings rising alarmingly fast in the past several years. Back in 2016, a marine biologist said “It’s not if but when, in terms of somebody being fatally attacked” in Cape Cod. “We’ve got seals being eaten within 100 meters of surfers.”And, unfortunately, he was right. On August 15thof this past summer, a 61-year old man was swimming just 10 feet off the shore of Truro Beach when he was attacked by a great white. He had to undergo six extensive surgeries, but he survived the attack. Then, just a few weeks later, on September 15th, 26-year old Arthur Medici was boogie boarding about 30 feet offshore in shallow water when he was fatally wounded when a shark bit down on both of his legs. This was the first fatality in the United States since 2015 and the first fatal incident in Massachusetts since 1936.

Arthur Medici’s memorial service this past fall.

Because of the explosion of the seal population on the cape in recent years, the great white problem on the cape is very new, and there is a lot of disagreement on what needs to be done to keep beachgoers safe. Although shark attacks are not extremely common, human shark interaction is only on the rise, meaning that this is not a problem that will be going away any time soon. There will always be people willing to take the risk and continue venturing into the water to swim and surf regardless of the possibility of an attack, so it is important to do all that we can to keep them safe. I was curious what places in other parts of the world where the threat of sharks is nothing new have been doing to deal with the threat, and if there could be an application for AI, and the answer is yes.

A group in Australia called The Ripper Group, in conjunction with The University of Technology Sydney, developed a system called SharkSpotter. SharkSpotter combines breakthrough artificial intelligence (leveraging deep-learning networks and image processing techniques), computing power, and drone technology to identify and alert lifesavers in real time to sharks near swimmers, thereby creating a cost-effective way to monitor beach safety over very large areas.

Developed using machine learning techniques known as “deep learning”, the SharkSpotter system receives streaming imagery from the drone camera and attempts to identify all objects in the scene. Once objects are detected, they are put into one of sixteen categories: shark, whale, dolphin, rays, different types of boats, surfers, and swimmers. If a shark is detected, SharkSpotter provides both a visual indication on the computer screen and an audible alert to the operator. The operator verifies the alert and sends text messages from the SharkSpotter system to the Life Savers on the beach for further action.In an emergency, the drone is equipped with a lifesaving flotation pod together with an electronic shark repellent that can be dropped into the water in cases where swimmers are in severe distress, trapped in a rip, or if there are sharks close by. Conventional techniques such as fixed-wing aircraft spotters and helicopters with human spotters have very low detection accuracy, about 12.5% and 17.1%, respectively. These advanced machine learning techniques significantly improve aerial detection to more than 90% accuracy. After successful trials and fine-tuning of the system, SharkSpotter was used across a dozen popular beaches in New South Wales and Queensland this past summer. The system was developed to help Surf Life Savers monitor the beach more effectively, as opposed to replacing them (as humans are still needed on site to perform any necessary first aid in the event of an incident), and has been received positively by end-users and communities alike, according to a survey recently conducted by The Ripper Group. In January 2018, SharkSpotter was successfully used to rescue two young swimmers caught in a rip tide in Australia. The drone identified the swimmers in distress, flew down the beach about 800 meters from the lifeguard station, and dropped a lifesaving flotation pod. The complete rescue operation took about 70 seconds, much faster than any human-centered effort would have been able to.

The benefits of using AI for Shark Detection can be mainly split into three categories: maintaining beach safety, protecting marine environments, and enhancing tourism.

As I’ve outlined before, the accuracy of using AI to monitor the water is astronomically better than any other methods out there, and as this can be a matter of life and death, any increased accuracy is important. The current most common method of tracking sharks is the use of tags. Scientists will track down sharks, attack a tag to them, and release them again. Because it is too expensive to constantly track them, the tags are connected to certain buoys and notify the team when a tagged shark swims in the vicinity of those buoys. This method is currently used on the Cape and connects with a free downloadable app called Sharktivity. The problem with this, however, is that it is unrealistic to think that all sharks in the vicinity can be tagged (there are well upwards of 100 sharks now, possibly much more), and it doesn’t allow sharks to be actively tracked, only when they are near certain checkpoints. 

One common method for shark protection, particularly in Australia and South Africa, is the use of shark nets. However, shark nets are controversial because they are designed to kill potentially dangerous sharks. In the process, nets may also injure or kill non-target animals, including endangered and protected species. If any of you watch Shark Week (highly recommend if you don’t), you’ll know that many kinds of sharks are very endangered and they are extremely vital to maintaining the marine ecosystem, so any way to monitor them while avoiding lethal action if possible is ideal.

Almost all of the places where shark attacks are a concern are also popular tourism destinations (The Cape, Australia), and the local businesses depend on the tourists to keep their community going. Already some business owners in The Cape are concerned for the well-being of their shops. People need to feel safe enough to continue going to and spending money at these destinations, and if the Cape wants to avoid turning into the real-life Amity Island, accurate shark detection is a big and important step in getting there.

Sources:

https://therippergroup.com/industry-solutions/shark-recogniiton/

https://www.floridamuseum.ufl.edu/shark-attacks/

https://www.nationalgeographic.com/animals/2018/11/wild-returns-cape-cod-shark-attacks/

https://www.youtube.com/watch?time_continue=17&v=V0p_XOqlZfk

https://theconversation.com/sharkspotter-combines-ai-and-drone-technology-to-spot-sharks-and-aid-swimmers-on-australian-beaches-92667

https://www.bostonmagazine.com/news/2013/06/25/gray-seal-population-problem-cape-cod/2/

7 comments

  1. Great post Shannon! Honestly SharkSpotter combines three of my favorite things, sharks, drones, and data analytics. This is a great application for how deep learning with images can save lives, similar to the Ted Talk we watched a couple of weeks back. Personally though, I’m surprised that they use a drone as the method of delivery and analysis, not only because of how short the battery life is, but the risk of overheating the motors. I agree shark nets are too cruel and buoys seem a little out dated and idealistic, SharkSpotter definitely fills the gap and I hope that if/when it comes to the Cape it gives people a little more piece of mind. Personally, I think that they machine learning efforts should be placed rather into investigating sharks behavior towards humans, as risky as that sounds. Sharks are largely misunderstand creatures and more research should go into understanding our interactions rather than just identifying them and running away. As a scuba diver I’ve had my fair share of shark encounters and in my experiences, they’re more scared of me than I am of them and they swim away. Great post!

  2. The detection accuracy statistics that you provided – 12-17.5% for traditional methods vs 90% for SharkSpotter – are staggering and drive home just how radically different machine learning applications can be from their predecessors. Even beyond the good it can do for shark attack prevention, I wonder if the particular combination of technologies that SharkSpotter leverages (drone-enabled, image-based deep learning) can be/is being used for other applications in the same marine life or environmental protection spheres. For example, I can envision a more advanced version of the SharkSpotter concept being used to track endangered species, to provide better coastal monitoring for national security or Coast Guard rescue purposes, or to more rapidly detect underwater oil pipe ruptures (most likely by the abnormal shape/color of the spill plume beneath the surface). Really interesting topic!

  3. Awesome post, and presentation last week – it’s such a unique topic! It’s so interesting to see that this new technology can be applied to radically improve the accuracy of shark spotting, and safety of beachgoers alike. I assume that the revenue model for the Ripper Group would target the municipalities themselves to pay for their services, and this seems like a great investment for . any town trying to improve tourism and keep people safe. As someone who spends a fair amount of time surfing in the area, it is exciting to hear of the new technologies that can keep us all safe!

  4. Awesome follow-up to an already interesting presentation. With the increasing accuracy of image detection, it makes perfect sense that this technology would be applied to spotting sharks and keeping beaches safe. Coming from lake country in the midwest, I have never really worried about sharks, and the increased activity on the Cape sounds like a very serious issue. The two technologies of drones and AI combined, prove to be a really impactful solution to such a dangerous problem. I think it’s awesome that the drones can even carry life-preservers and shark repellent, making response times substantially lower than able to be provided by just humans on the beach. This seems like an awesome solution not only to shark control, but as you noted looking out for distressed swimmers or any potential dangerous situations on the beach. Great job on both the post and presentation, I learned something new!

  5. I really appreciate this and it hits on all the things I love about Massachusets. The beach, ocean, and sharks. As an avid shark fisherman, beachgoer and boat owner I have a pretty good idea of the problems at stake, and several of the “solutions” that have been presented. Of all of them, this is by far the best and I feel a partnership with AIRA insights PARC drone would be a good match as it would remove the downtime and charging time for the drones, plus they fly without an operator and are more powerful because they are tethered to the ground. That being said, I hope those on the Cape keep a sharp lookout this summer, the more we continue to cohabitate in waters occupied by shark dinner, the more issues will arise.

  6. This is a great follow-up to your presentation last week, and I think it’s such a unique application of AI to help solve a problem we are facing on our own beaches. The stats you referenced comparing accuracy among different types of detection are staggering – clearly, there is a need for AI in this context when it can improve detection accuracy by such a large percentage. The economic angle is interesting too – if local business want to keep tourism rates up in these popular vacation areas, they might feel inclined to help fund these AI programs. In that context, it seems like everyone would win.

  7. Such an interesting presentation topic and a great post to wrap it up. I honestly had no idea how bad the shark problem was in cape cod !! As much as that thought disturbs me though I love the application of technology to solve it. The improvement in detection statistics is amazing and I think the potential it has for creating safe beaches that are also marine life friendly is what I’m most excited about ! Thanks for highlighting such a fascinating topic !

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: