Critically endangered curlew sandpipers roosting at Bayside Park, Manly in Brisbane.
(Photo credit: Chris Walker
)
Drones and artificial intelligence could provide valuable information to guide efforts to slow the alarming extinction rate of birds across the globe, research from The University of Queensland has found.
A UQ-led global study has found AI can accurately detect birds in drone imagery 85 per cent faster than humans, improving access to and understanding of birds in need of effective conservation.
Study coauthor Professor Richard Fuller from UQ’s School of the Environment said hard to reach locations often prevented the effective population monitoring required to guide conservation efforts.
“Birds are going extinct at alarming rates, similar to those seen during the 5 major mass extinction events in Earth’s history,” Professor Fuller said.
“Effective conservation depends on fast and scalable monitoring methods.
“This study lays the foundation for using AI to efficiently process drone bird survey imagery across vast and difficult landscapes.”
A global dataset containing almost 50,000 birds from more than 100 species was used to train and test the AI.
Researcher Cesar Fernandez used a drone to monitor breeding populations of Chilean flamingos in southern Peru.
(Photo credit: Cesar Fernandez)
The AI identifying the birds in the drone photo.
(Photo credit: Cesar Fernandez)
Lead author Dr Joshua Wilson said the study result could improve monitoring in the hundreds of Australian wetlands inhabited by thousands of species, including the critically endangered Curlew Sandpiper.
“Manually counting birds in drone imagery is extremely time-consuming, and this study shows that AI can reduce this burden,” Dr Wilson said.
“With AI offering a way to process large numbers of images efficiently, drone surveys can be carried out more often and at larger scales.
Dr Joshua Wilson flying a drone near a wetland critical to shorebirds.
(Photo credit: The University of Queensland)
“Drones and AI aren’t the right tool for every situation, but they work particularly well for large-bodied birds in open habitats with little canopy cover, like wetlands.
“In these cases, drones and AI shouldn’t replace experienced surveyors, they should free them from wading through knee-deep mud or manually detecting every bird in tens of thousands of images.
“This gives experts more time to interpret results, design effective monitoring programs and advocate for conservation action.”
The drone-based bird dataset is now the most accurately labelled and biologically, environmentally and digitally diverse collection of its kind.
Researchers hope it lays the foundation for future research.
The trained model is freely available for use, while the open-access dataset provides the foundation for future research and the development of improved AI systems.
More than 30 researchers from 11 countries contributed images to strengthen the AI identification process and helped construct the paper published in Remoting Sensing in Ecology and Conservation.
