Studying coral reefs used to mean hours of painstaking manual analysis, but artificial intelligence is changing the game. A new neural network can process ocean sounds in real time and identify fish activity 25 times faster than humans. This technology will revolutionize the way scientists monitor coral reef health and protect marine ecosystems.
Coral reefs are among the most diverse ecosystems on Earth. Although coral reefs cover less than 1% of the ocean, they provide habitat for approximately 25% of marine species at some stage in their life cycles. With so much biodiversity concentrated in one place, scientists face challenges in identifying exactly which species and how many.
To solve this problem, researchers at Woods Hole Oceanographic Institution have developed a new method that combines acoustic monitoring with neural networks to analyze fish activity based on sound. Their research report was published today (March 11) in JASA, the journal of the Acoustical Society of America, published by AIP Press.
For years, scientists have relied on passive acoustic monitoring to study coral reefs. This involves placing an underwater recorder on the reef for several months to capture environmental sounds. While existing signal processing tools can analyze large amounts of audio data, they are not designed to detect specific sounds. To identify individual fish calls or species-specific sounds, researchers still have to manually sift through hours of recordings.
Author Seth McCammon said: "Honestly, it's a horrible job for the people who do it. It's incredibly tedious work. It's so painful."
Equally important, this manual analysis is too slow for practical applications. Many of the world's coral reefs are threatened by climate change and human activities, so being able to quickly identify and track changes in coral reef populations is critical for conservation efforts.
"It took humans years to analyze data to this extent. Analyzing data in this way just doesn't work at scale," McCammon said.
As an alternative, the researchers trained a neural network to automatically sort through large amounts of acoustic data, analyzing audio recordings in real time. Their algorithm is as accurate as human experts at deciphering coral reef acoustic trends but is more than 25 times faster, and it could change the way oceans are monitored and studied.
"Now that we no longer need humans involved, what other types of devices can we use besides recorders? Some of the work my co-author Aran Mooney is doing includes integrating this neural network onto a floating mooring that will update fish sound counts in real time. We are also working on getting our neural network onto our autonomous underwater vehicle, CUREE, so it can listen to fish sounds and map out hotspots of biological activity," McCammon said.
The technology also has the potential to solve a long-standing problem in ocean acoustics research: matching each unique sound to a fish.
"For the vast majority of species, we're not yet at the point where we can say with certainty that a certain sound is coming from a certain fish," McCammon said. "That, at least to me, is the Holy Grail we're looking for. By detecting fish sounds in real time, we can start to build devices that can automatically hear the sounds and then see what fish are nearby."
McCammon hopes that such neural networks will eventually provide researchers with the ability to monitor fish populations in real time, identify problem species and respond to disasters. At a time when coral reefs need all the help they can get, this technology will help conservationists get a clearer picture of their health.
Compiled from /ScitechDaily