University of Maine researchers are teaming up to create artificial intelligence that can identify and count birds in aerial photos. The technology could help biologists learn more about bird populations, migrations and behaviors.
“This project provides an exciting opportunity for wildlife and computer science students to work together to apply emerging technologies to help wildlife conservation practitioners solve a real-world problem,” said Cynthia Loftin, associate professor of wildlife ecology and leader of the United States Geological Survey Maine Cooperative Fish and Wildlife Research Unit.
The project recently received $43,000 from the UMaine AI Initiative seed grant funding program, and builds on previous grants and partnerships involving UMaine faculty and state and federal agency partners.
The project coincides with the UMaine AI Initiative, an effort to transform the state into a world-class hub for artificial intelligence research and education.
“This is the future of a lot of research, interdisciplinary teams that also involve AI analyzing data,” said Roy Turner, associate professor of computer science and director of the Maine Software Agents/Artificial Intelligence Laboratory. “I think this is going to be the normal kind of way of doing business, interdisciplinary teams that usually involve some sort of computer science.”
Faculty and graduate students from several departments at UMaine will collaborate in the project, developing technology that can pinpoint colonial nesting birds in photos captured by cameras mounted in unmanned aerial vehicles or planes. The AI developed will use object recognition and image segmentation to determine the number of birds, their species and behaviors in photos captured on Maine’s offshore islands and over inland rookeries.
For years, researchers have manually counted and identified birds in aerial photos, a process that consumed many hours.
“Humans are prone to fatigue, error,” Turner said. “It takes forever to do this by hand. Graduate students can take several hours identifying birds in one image.”
Turner and his team plan to develop their Convolutional Neural Network, a deep learning AI algorithm typically used for visual analysis, using a method for image segmentation called Mask R-CNN, although Turner says they will explore other tactics. The network will find and classify the birds in an image by analyzing the pixels that form them. Turner says the network analyzes the pixels in much the same way a person’s visual system does to detect and identify objects.
The technology could provide scientists more information about birds in inaccessible areas like remote islands with dangerous terrain and rookeries with birds nesting in the canopy tops, Loftin says. The new method of surveying could also reduce disturbance to colonies, particularly by eliminating the need to walk around in their habitats.
Researchers involved in the effort include Turner; Loftin; Salimeh Yasaei Sekeh, assistant professor of computer science; Kate Beard-Tisdale, professor of spatial information science and engineering; Daniel Hayes, Barbara Wheatland Associate Professor of Geospatial Analysis and Remote Sensing; David Sandilands, aerial survey pilot and remote sensing technician with the Wheatland Geospatial Lab in the School of Forest Resources; and Anthony Guay, remote sensing technician specialist with the Wheatland lab and School of Forest Resources.
Professors recruited Alex Revello, a master’s student of computer science, to help integrate the CNN. Meredith Lewis and Logan Kline, master’s students in the Ecology and Environmental Sciences program in the Department of Wildlife, Fisheries, and Conservation Biology, are developing protocols for using UAVs to collect imagery of nesting colonial birds and evaluating how this methodology can reduce disturbance while also enhancing survey efficiency. The team also employed undergraduates in summer 2020 fieldwork and recruited a new group of seniors to help with the project this semester and in spring 2021 as their Capstone project.
To test their network, the scientists will task it with pinpointing and specifying birds in images that other researchers have already analyzed to ensure the AI achieves the same results. Turner says his team plans to develop, evaluate and launch the machine learning tool by fall 2021.