New tagless fish detection technology based on Artificial Intelligence has been developed by Innovasea to help bring the task of fish counting into the 21st century. IWP&DC discovers how this innovative technology could be a key tool in a dam and hydropower plant operator’s toolbox
Aerial view of White Rock Dam
The monitoring of fish activity around hydropower and dam facilities can be a tedious task. Employees often have to watch hours of footage recorded by underwater cameras to manually count fish on the screen and attempt to identify the species they see. This time-consuming and often inefficient process could however soon become a thing of the past thanks to the development of a new tagless fish detection technology that can detect, count and classify fish in real-time using a combination of optical cameras, imaging sonar and artificial intelligence.
The innovative technology – developed by Innovasea in conjunction with DeepSense and Nova Scotia Power as part of the Ocean Aware project, which is partially funded by Canada’s Ocean Supercluster – has been shown in field trials to be more than 95% accurate in properly counting and identifying wild fish. Importantly, the testing has also indicated that the technology could provide a deeper understanding of how fish behave around hydro facilities so operators can both lessen their impact on wildlife and reduce the number of days they shut down or modify operations to mitigate those potential impacts.
Testing times
The tagless system was initially unveiled back in May 2022, following successful preliminary testing that was performed using video of fish swimming around the Wells Dam in Washington’s Columbia River along with proprietary data sets from Emera and high-resolution DIDSON imaging sonar captured from the Ocqueoc River in Michigan. This early testing showed the system was 90% accurate in properly counting and identifying wild fish. In Spring 2022, field tests were carried out for the first time at the White Rock Dam on the Gaspereau River in northern Nova Scotia. The technology was used to monitor the migration of the enormous schools of alewife that passed the 3.4MW generating project. Over previous years, the project relied on manual counting from recorded video to determine when the migration activity has slowed down.
For the field test, one optical camera was installed at the top of the fish ladder, and the other in the fish bypass. Two sonar cameras were installed upstream and downstream from the dam. Once the system was up and running, data was sent to the cloud and could be accessed in real-time via an Innovasea app that showed the rate of fishing passing by each minute, the cumulative total of fish and a live camera image that gets refreshed every 60 seconds.
In all, the AI-powered system, which monitored the busy spawning period between mid-April and late June, counted nearly a million fish – 918,169 to be exact. Most interesting, the data indicated two distinct phases of the alewife migration.
In the first phase, which ended around May 18, the sonar and IP cameras discovered that all the fish were going up the fish ladder and turning left to swim upstream to spawn. In the second phase, the fish were more erratic, swimming all over the place. In addition, the data showed that the alewife prefers being on the move during the daytime.
“The data resolution we got was amazing,” explained Jean Quirion, Innovasea’s vice president of research and development for fish tracking. “Before this people were manually counting fish from recorded video, and there’s no way they could produce this same level of data because they don’t have time to watch video 24/7 and notice every fish.”
The final results from the AI system were validated against the results from manual counting. During 178 random samplings, Innovasea personnel counted 2,564 fish from recorded video while the tagless detection system counted 2,681, a difference of less than 5 percent.
“We’ve had phenomenal results,” said Quirion. “Nobody has ever recorded this kind of data before. With the traditional approach of manually counting from a video, there is a fatigue aspect to it, there’s human error etc. With this new AI system, a machine does it automatically 24/7. There’s no limit to how many cameras it can watch – this is something that was never before possible. What’s exciting is to see the amount and quality of data that’s delivered, all in real-time.”
These kinds of insights, coming in real-time straight to the end user on a mobile phone or a web app, hold the potential to make hydropower facilities more precise in how they operate – halting power generation when fish are nearby but minimizing downtime when the fish aren’t active. Rather than shut down for days or weeks, for example, operators could run overnight and stop at dawn when fish are on the move again.
Quirion explained: “At White Rock last year, we had one camera looking at fish coming up the fish ladder. The whole migration count is about 900,000 fish. And we know that we’ve validated the system to have a 95% accuracy. In addition to that, the data that the system provided had minute-level resolution. Every minute we get a new count of how many fish pass by and we can clearly see movement. We were seeing patterns throughout the day. Previously it would not have been possible to differentiate between these patterns with that level of resolution.
“In the future, decision-making for power generation may be enabled by this technology,” suggested Quirion. “The patterns we are seeing, are daily, but also offers the opportunity to see seasonal patterns. The system currently provides minute-level resolution, and we’re working on reducing that to even to less than minute resolution. What kind of operational decision might be enabled by a capability like that?”
Additional trials and investigations
While the results from White Rock have been promising, Innovasea is looking to perform additional field trials of the system with other utilities or dam operators, including in countries outside of North America.
“The more sites we have access to the more versatile this technology is going to become,” said Quirion. “It would also be interesting to look at sites in different countries. There might be different problems in Europe, for example, than North America. The more variety we have, the better.”
Of particular interest in the next round of trials is seeing how the AI system performs in identifying different species of fish. That’s something it wasn’t required to do at the White Rock facility, which sees almost exclusively alewife using its fish passage systems.
The company is also interested in investigating the use of the technology at larger hydropower facilities. “There is the range limitation on the cameras and sonars we use. We could potentially combine cameras to be able to see at a longer range,” said Quirion. “We’d love to have a partner that has wider canals so we can explore these possibilities.”
“We also want to explore the benefits of the technology in different contexts,” Quirion added. “We have augmented this new system at White Rock with our traditional acoustic telemetry and fish tags too. We’ve set up receivers at various points in the channel or the river, allowing us to add a second layer of data. Near the end of the year, once we got the results from the initial study, we were able to compare them with our AI system, and there was some very good agreement there. Our traditional tag technology is really good at tracking individual fish, so you can see the route and behaviour of individual fish, whereas the AI system can’t track individuals but it can see the masses. It sees the whole population and gives a much greater statistical sample. It was great to compare these two technologies. We’re aiming to provide a different tool, a new tool in the toolbox that never existed before. One that gives advantages that are complementary to our traditional tagging technology.”
For now, the company is planning to repeat the tests it did last year at the same site and analyze the results. That will be a big milestone to confirm that the system is seeing the same movement patterns, for example, throughout the migration of the fish.”
Ocean Aware project
Canada’s Ocean Supercluster announced its Ocean Aware project back in 2020. The Innovasea-led project is a $29 million program to develop new technologies to help maritime industries improve operations while lessening their impacts on the environment.
Ocean Aware will deliver benefits in four key areas:
- Aquaculture – Technological advances and new tools will give fish farmers an enhanced ability to monitor environmental conditions and protect fish from harmful underwater events.
- The Fishing Industry – New technology to observe, monitor and locate fish stocks will help increase fishing efficiency, improve the sustainability of our fisheries and minimize bycatch.
- The Energy Sector – New tools to better monitor and understand the behaviour of the marine life that tends to congregate around fixed infrastructure in the ocean, such as offshore wind turbines and oil and gas platforms.
- Ocean Discovery/Research – New sources of data for scientists and researchers so they can conduct ocean ecosystem research and gain a deeper understanding of marine life movement and activity.
While a Canadian initiative, Ocean Aware will create a blueprint – and the tools and technologies – that other nations can use to ensure that important maritime industries operate in a way that is both sustainable and profitable.
Working with Innovasea is an impressive team of partners, supporters and stakeholders who each bring a wealth of expertise and insight to the table. They include Emera, Nova Scotia Power, Ocean Choice International, Irving Shipbuilding, Dartmouth Ocean Technologies, Xeos Technologies, and with the support of the Ocean Tracking Network, Dalhousie University, Fisheries Marine Institute of Memorial University, the Department of Fisheries and Oceans, and others.
In addition to leading this impressive collection of experts, Innovasea’s role will also be to build upon the advanced fish tracking and aquaculture intelligence solutions already developed and adapt them as needed based on the requirements of the project.
This article first appeared in International Water Power magazine.