Sara Beery came to myth as an assistant professor in the Department of Electrical Engineering and Computer Science MIT (EECS), willingly focusing on ecological challenges. She changed her research career regarding the possibility of applying her specialist knowledge in the field of computer vision, machine learning and data learning to solve real problems related to protection and sustainable development. Beery attracted the Institute's involvement in “Calculation of the planet” and decided to introduce her methods to global monitoring of the environment and biological diversity.
In the north -western Pacific, salmon has a disproportionate impact on the health of its ecosystems, and their complex reproductive needs attracted Beere's attention. Each year, millions of salmon begin migration to revival. Their journey begins in freshwater beds, where the eggs hatch. Young salmon (newly hammered salmon) go to the ocean, where they spend several years maturing to adulthood. As adults, salmon returns to streams in which they were born to be reborn, ensuring the continuation of their species by depositing eggs in the gravel of stream beds. Both male and female salmon die shortly after delivering the habitat of the river with the next generation of salmon.
During their migration, salmon supports a wide range of organisms in ecosystems that they undergo. For example, salmon brings nutrients such as carbon and nitrogen from the ocean, increasing their availability to these ecosystems. In addition, salmon is crucial for many predators-appointments: they serve as a source of food for various predators, such as bears, wolves and birds, while helping to control other populations, such as insects, through predation. After death, due to spawning, the decaying corpses of salmon also complement valuable nutrients to the surrounding ecosystem. Salmon migration not only maintains its own species, but plays a key role in the overall health of rivers and oceans that they live.
At the same time, salmon populations play an important role both economically and culturally in the region. Fisheries of commercial and recreational salmon significantly contribute to the local economy. And for many indigenous people in the northeastern Pacific, salmon has a significant cultural value because they were crucial for their diet, tradition and ceremonies.
Monitoring of salmon migration
Increased human activity, including the transition and development of water energy, along with the loss of habitats and climate change, had a significant impact on salmon populations in the region. As a result, effective monitoring and management of salmon fisheries is important to ensure a balance between competitive ecological, cultural and human interests. Accurate counting of salmon during seasonal migration to their Natal River for Rebirth is necessary to track endangered populations, assess the success of recovery strategy, conducting the regulations of the fishing season and support the management of commercial and recreational fisheries. Accurate population data help decision -makers use the best strategies to protect the ecosystem health while satisfying human needs. Monitoring of salmon migration is an intense and inefficient undertaking.
Beery is currently conducting a research project, which aims to improve salmon monitoring using the most modern methods of computer vision. This project is in a broader research interest in Beery, which focuses on the interdisciplinary space between artificial intelligence, the world of nature and sustainable development. His importance for fishing management meant that it was good for financing Abdul Latif Jameel Water and Food Systems Lab (J-Wafs). Grant Seed Grant Beery 2023 J-Wafs was the first financing of the research she received since joining the MIT Faculty.
Historically, monitoring of efforts consisted of people to manually count the salmon from the edges of the rivers with sight. Over the past few decades, underwater sonar systems have been implemented to help count the salmon. These sonar systems are basically underwater video cameras, but they differ in that they use acoustics instead of light sensors to capture the presence of fish. The use of this method requires people to set up a tent next to the river to count the salmon based on the output of a sonar camera connected to the laptop. Although this system is an improvement in the original method of monitoring salmon through sight, it still depends significantly on human effort and is a tedious and time -consuming process.
Automation of salmon monitoring is necessary for better management of salmon fishing. “We need these technological tools,” says Beery. “We cannot keep up with the request of monitoring and understanding and studying these really complex ecosystems in which we work without any form of automation.”
To automate the counting of the migration of salmon population on the north -western Pacific, a design team, including Justin Kay, a PhD student at ECS, collects data in the form of films from Sonar cameras in various rivers. The team adopts a subset of data to train the vision system for autonomous detection and counting of fish during migration. Kay describes the process in which the model counts every migrating fish: “The computer vision algorithm is designed to locate the fish in the frame, draw a box around it, and then track it over time. If the fish is detected on one side of the screen and leaves it on the other side of the screen, then we count it as moving up.” On rivers where the team created training data for the system, he brought strong results, with only 3 to 5 percent of the error. This is much below the purpose that the team and partners set no more than 10 percent counting error.
Testing and implementation: balancing human effort and use of automation
Scientists technology is implemented to monitor the migration of salmon on the re -restored Klamath River. Four dams on the river have recently been demolished, which makes it the largest project to remove there in US history. The dams collapsed after over a 20-year campaign to remove them, run by Klamath tribes, in cooperation with scientists, environmental organizations and commercial fishermen. After removing the dams of 240 miles of the river now flows freely and almost 800 square miles of habitats are available to salmon. Beery notices almost immediate regeneration of the salmon population in the Klamath river: “I think that within eight days of stopping the dams they began to see how salmon actually migrates up the river beyond the dam.” In cooperation with California Trout, the team currently processes new data to adapt and create a tailored model, which can then be implemented to help count the newly migrating salmon.
One challenge related to the system rotates around the model's training to accurately count the fish in unknown environments with variants, such as the features of the riverbed, water transparency and lighting conditions. These factors can significantly change the way fish appear at the output of the Sonar Camera and confuse the computer model. After implementation in new rivers, where no data was collected before, as well as Klamath, the system performance degrades and the margin of errors significantly increases to 15-20 percent.
Scientists have constructed an automatic adaptation algorithm in the system to overcome this challenge and create a scalable system that can be implemented anywhere without human intervention. This self -control technology works to automatically calibrate to new conditions and the environment to accurately count migrating fish. When testing, the automatic adaptation algorithm was able to reduce the counting error to the range from 10 to 15 percent. Improving the counting error in the function of self -control means that technology is closer to implementation in new locations without much additional human effort.
Enabling real -time management using “Fishbox”
Another challenge facing the research team was the development of efficient data infrastructure. To run a computer vision system, video produced by Sonar cameras must be delivered using a cloud or by manually sending hard disks from the river site to the laboratory. These methods have remarkable disadvantages: cloud -based approach is limited due to the lack of internet communication in distant locations of river sites, and data shipping introduces delay problems.
Instead of relying on these methods, the team implemented the energy computer, invented “Fishbox”, which can be used in the field to make processing. Fishbox consists of a small, lightweight computer with optimized software, which fisheries managers can connect to existing Sonar laptops and cameras. The system is then capable of launching salmon counting models directly in Sonar sites without the need for internet communication. This allows managers to make hourly decisions, supporting more responsive management of salmon populations in real time.
Community development
The team is also working on connecting the community in monitoring the fishing management of salmon on the north -western Pacific. “It is simply exciting that stakeholders will enthusiastically gain access to (our technology), because we will lead him to work and stronger integration and cooperation with them,” says Beery. “I think that especially when you work on food and water systems, you need direct cooperation to facilitate impact, because you ensure that what you develop actually satisfy the needs of people and organizations you help.”
In June, the Beery laboratory organized workshops in Seattle, which convened non -governmental organizations, tribes as well as the state and federal departments of fish and wildlife to discuss the use of automatic sonar systems for monitoring and managing salmon populations and managing them. Kay notes that the workshop was “an amazing opportunity for everyone to divide the different ways in which they use Sonar, and wondering how the methods we build, we build, can fit in this flow of work.” The discussion now lasts through the joint Slack channel created by the team, with over 50 participants. A supporter of this group is a significant achievement, because many of these organizations would not have any opportunities to meet and cooperate.
Waiting for something
When the team continues to tune the computer vision system, improve their technology and cooperates with various interested parties – from the indigenous communities to fisheries managers – the project is able to introduce a significant improvement in the efficiency and accuracy of monitoring and management of salmon in the region. And because Beery develops the work of his MIT group, the J-Wafs seed grant helps to keep challenges, such as fishing management on its sight.
“The fact that the J-Wafs grant existed here in myth allowed us to continue working on this project when we moved here,” Beery comments, adding “he also expanded the scope of the project and allowed us to maintain active cooperation on what I think is a really important and influential project.”
When J-Wafs means the 10th anniversary this year, the program aims to continue supporting and encouraging MiT lecturers to implement innovative projects that aim to develop knowledge and create practical solutions with real impact on global challenges of the water and food system.
