University of Missouri scientists are testing out a new way to count waterfowl — using drones and artificial intelligence.
The Missouri Department of Conservation conducts weekly surveys of waterfowl to understand how many birds are in the state and inform management practices.
Staff members do the majority of these surveys from the ground and also conduct some from airplanes. Right now, there’s not really a standard practice for conducting the surveys, said Reid Viegut, migratory game bird coordinator with the department.
Site managers “all have kind of learned from each other and picked up different tricks and techniques to use to make their counts better over the years,” Viegut said.
About eight years ago, the department started talking about using drones to do the surveys with Yi Shang, a computer science professor at the University of Missouri who has worked on technology projects with the department for 15 years. Shang wanted to find a way to improve the process.
“People, just based on their eyesight, estimate how many birds are there, and scheduling fixed-wing airplanes has been a quite difficult, expensive process,” Shang said. “With drones, we think that would be a much-preferred technology that can offer low cost, cost-effective and also faster image processing based on the advances in AI technology.”
Shang led this new study, testing methods of counting waterfowl with drone imagery and AI at 10 sites around the state. The project was funded by the Conservation Departament.
Students flew drones over geese, ducks or other waterfowl, taking photos of the birds. They then used AI software to merge the images into a mosaic, count the different types of birds and identify the type of habitat. They also made sure they were not double-counting birds. Finally, the scientists used ChatGPT to create an automatic report of the survey.
“Before that, MDC collected lots of images — they don't have the manpower even to process them,” Shang said with a laugh. “If you just look at the picture from the drawing, you only see those small dots, you have to zoom in, so this is a very tedious and time-consuming process.”
Viegut knows that firsthand. Before taking his current position, he worked on the study with Shang and served as a human baseline, zooming in on photos and counting ducks to understand how accurate the AI system was. The department isn’t using the new technology quite yet, but this is a job Viegut hopes AI takes.
“I would never envy anyone who said their job was to look at images and label ducks, especially after doing hundreds of thousands of them,” Viegut said.
The system also wouldn’t replace the person conducting the survey because the drone and the technology would still need to be operated by a person, Viegut said.
Shang and his team found the AI system was able to correctly identify ducks on easy backgrounds like water at over 95% accuracy. When they were in a harder habitat, like vegetation, the accuracy rate fell but was still over 80%.
Viegut said more accurate counts of migrating birds will help the state make better decisions about management, and his coworkers are excited about the potential.
“I don't think that it's going to end up taking anything away from area managers,” Viegut said. “They're all looking forward to it. They're really excited about the project.”
Viegut also studied how the waterfowl reacted to the drones and did not find a change in their behavior.
“That's really positive because when we're driving around to count these ducks, occasionally we are disturbing them, or in the planes, when we fly those surveys, we're occasionally disturbing them,” Viegut said.
The study on waterfowl and AI was published in the journal Drones. Shang is working on other projects with the department that apply image recognition to wildlife. One involves identifying animals that are seen on camera traps across Missouri. Another identifies the age of a fish based on a section of its bone.
“Without processing this data, you don't have the information and knowledge,” Shang said. “So AI is the way to do it. Otherwise, I don't think we will be able to take advantage of the amount of data we have.”
