AI software developed in N.S. provides glimpse into wild and murky Bay of Fundy
Facial recognition software has been adapted to filter echo sounding data
Artificial intelligence software developed in Nova Scotia is making it easier and faster to detect marine life in the turbulent and cloudy waters of the Bay of Fundy — a milestone in the effort to monitor the potential impact of tidal turbines.
The ferocious tides that have so far thwarted efforts to generate renewable electricity on a large scale also make it difficult to get a picture of what's down there.
The Bay of Fundy is so murky with churned sediment, researchers use elaborate fish finders that emit echo locating pings to "see" what's swimming in the water column.
But the tides also generate huge amounts of air bubbles that obscure fish and other marine life.
"That causes a lot of problems because we need to sort out what is fish and what is air bubbles so that we're not counting air bubbles as biological signals," said Jessica Douglas, an ocean technologist with the non-profit Fundy Ocean Research Centre for Energy, or FORCE.
Hard to see down there
Her job is to monitor marine life in the Minas Passage where a demonstration site has been created to test tidal power technology.
The first turbine lowered to the bottom was destroyed by the tides within days. The second was abandoned by a bankrupt developer and remains on the bottom. Still, other turbines are expected to be deployed this year.
Until recently, Douglas used a computer mouse to manually draw lines around the air bubbles — a time-consuming process that could take 70 to 80 hours to scrub 24 hours of acoustic monitoring data.
"It's very tedious to make all those very small and minute edits, and it was taking me weeks to edit a data set."
That's where the ocean-focused DeepSense team at the Dalhousie University computer science department came in.
Using completed data sets from Douglas, they trained a computer to automatically filter out the "huge fuzzy mess," said assistant professor Chris Whidden.
"It was quite a large challenge. It was our first time working with this specific kind of acoustic data, audio data for looking at things underwater."
DeepSense adapted the same model used in facial recognition software.
"We trained that to use this different kind of data where we're treating these acoustic data as an image. So it's trying to focus on what are the important parts," said Whidden.
Echofilter program sifts the data
The Echofilter software, as it's called, does not remove the human element entirely.
"It's something that I can run overnight," said Douglas. "I'll run it on the data while I'm sleeping and then I can wake up and just double-check that everything worked OK. It saves me more than half the time."
Whidden calls it a "perfect use" of artificial intelligence technology where automation is checked by people.
"Everything that we do with our machine learning algorithm is still fed into a user interface where it can be checked over carefully, so you don't get the same issues you get with projects where we're doing machine learning detection of people and certain groups are left out because they weren't in the training set. We're just saving a lot of time and helping people."
Near real-time monitoring
Proponents of tidal energy say speedier processing of echo sounding data will provide Canadian regulators with the near real-time monitoring they want in place when granting permits for tidal turbines.
"This is a milestone event. People do want to know that the tidal energy devices don't harm sea life, including fish and marine mammals," said Russell Dmytriw, research director at the non-profit Offshore Energy Research Association.
The Echofilter software is part of the association's Pathway program where more than a dozen technologies are being developed to help tidal power companies satisfy the Department of Fisheries and Oceans and other government oversight agencies.
"The Pathway program will allow technology developers to depend on a reliable environmental effects monitoring suite of instruments that's been regulator-approved and able to demonstrate in near real time the minimal effects on submarine biota, including fish and marine mammals, and in the Minas Passage," said Dmytriw.
Pushing equipment to the limit
The monitoring project has been an eye-opener for Douglas.
The fast-moving tides are incredibly hard on monitoring equipment designed for calmer waters in the open ocean.
"We're putting them down at five metres per second, sandblasting, and ... a lot of them come out damaged," she said. "Some of them come out better than others. But we're really putting them to the test."
Douglas is amazed that wildlife thrives in what she calls "a wild environment. The craziest place ever."
"I can't believe anything's down there, to be honest, when I look at the data and how turbulent it is," she said. "And I think it's incredible how marine animals are even using it at all. And there's just so many different species.
"That's one of the really cool things that we've discovered from doing all this data collection, is just how many species really are out there using this environment on a daily basis. And I think it's been more shocking than anything."