Images shot by a drone and using AI to recognize sharks in Long Beach, CA Credit: Courtesy|Shark Lab CSU Long Beach

Sept. 2020 marked the end to one of Cal Poly’s summer undergraduate research programs (SURP) where students and Cal Poly alumni developed a safer and cheaper tool to detect sharks off the coast of California using artificial intelligence.

Before the new technology, volunteers for the shark lab would have to comb through hours of drone footage frame by frame in order to spot a shark. With the group’s contributions, the team was able to improve the accuracy and speed of the process of the artificial intelligence which can now spot not only sharks but humans, seals, boats, dolphins and surfers.

During the three month long program, the group had to find an artificial intelligence that was accessible to all members, and train it in order to improve its accuracy for spotting sharks. Computer science senior, Kathir Gounder was in charge of implementing the artificial intelligence and teaching it to the group members.

“The funny thing about artificial intelligence is that for every problem, there are a gajillion models that you could use,” Gounder said. 

Because it is more about experimentation than theory, no one has set a mathematical theory to describe some of these models, Gounder said. She spent the first few weeks picking the right model that everyone in the group can understand and interact with.

Computer science junior Grace Nolan contributed to the project by training the artificial intelligence. 

“Now, the best way to train the AI is to show it a variety of different conditions so that it can improve accuracy,” Nolan said.

It took the entire group as well as the lab at CSU Long Beach about 36 hours throughout the summer to label about 3,800 images for the project, according to Nolan. Each photo is labeled by hand.

Nolan, who has grown up with a passion for scuba diving and the ocean said he was happy to find a crossover between computer science and marine life.

In addition to identifying the photos, Nolan worked on the front end of the software, meaning she developed the program that the biologists use and interact with. Nolan hopes to develop an app that marine biologists can use by next quarter.

Cal Poly graduate Caroline Skae brought a focus on marine science to the project and the hope that people will continue their research to learn more about shark’s behavior patterns when they are around people by graphing the sharks’ pattern of movement and speed.

Using a few videos of shark activity, Skae and the team can extract all of the data they are interested in, such as the shark’s exact GPS location, an estimation of its size and behavioral patterns, and then store all of the data to help scientists. 

“You don’t need to have a science degree, you can be a part of collecting scientific data,” Skae said.

Traditional methods of studying sharks involve physically capturing and tagging sharks that encroach upon their habits. Using the drone to estimate their size and proximity to humans in the oceans, the sharks can be left alone.

The technology is cheaper than other shark spotting projects because the artificial intelligence allows for the sharks to be recognized with cheaper drones rather than high-quality commercial ones, according to Skae.

Skae believes that their technology could help to protect sharks, as well as humans. Sharks are apex predators who help keep the ocean clean by eating sick and dying organisms in the ocean, according to Skae.

“They’re very important to the ecosystem and they’re also definitely under threat as well because we kill worldwide over 100 million sharks a year for their fins or in bycatch [when animals are unintentionally caught by fishermen trying to catch other fish] so that it’s really kind of scary because they’re really important to the ecosystem,” Skae said.

Though Gounder does not have a similar background in marine biology, he was interested in bringing computer science to other fields of study.

“I’ve always been interested in interdisciplinary stuff because I feel like right now we over specialize a little bit,” Gounder said. “ There shouldn’t be anything stopping art students from walking down the road to the computer science building to help out.”

The group managed to work well together despite the roadblocks that the pandemic put in place.

“In computer science, a lot of the products that they have you do are really independent, really individual,” Nolan said. “So one of the really cool things — I want to think kind of the harder thing was — being able to work really well and really closely with this many people.” 

However, for Gounder, finding the interdisciplinary connection between each member of the group was a challenge he hadn’t experienced before.

“We all have very diverse skill sets, so it’s like, ‘Okay, how do we divvy up the work perfectly so that we all can contribute?’ Gounder said.

Normally, for a project such as this, Gounder would do the project by pair programming, which is when two programmers work together to check each other’s code. Due to social distancing guidelines, the group had to check their code with each other by adding it to an online platform called GitHub, which Gounder describes as “Google Docs for code”  in order to review it.

Though the program has ended, Gounder has continued to work on the project as his senior research project along with other volunteers and alumni and the groups continue to keep in touch.

Nolan hopes to meet with the team in person next quarter to fly a drone for themselves.

Leave a comment

Your email address will not be published. Required fields are marked *