Images captured by University of Queensland wildlife monitoring cameras.
The power of AI has been harnessed to rapidly clear a photography bottleneck and bring greater coordination and computing power to efforts to save Australian animals from extinction.
Developed by researchers at The University of Queensland, the Wildlife Observatory of Australia (WildObs) can quickly analyse millions of images taken by hidden wildlife cameras, meaning faster, more accurate data to guide conservation work.
Associate Professor Matthew Luskin from UQ's School of the Environment said the new cloud-based, easy to use, artificial intelligence-powered image platform was revolutionary.
"Affordable cameras can discreetly capture wildlife while strapped to trees and left for months so there are now thousands of projects across Australia collecting millions of images and videos," Dr Luskin said.
"We have unprecedented visibility into the natural world, but we were struggling to turn that information into timely, actionable data and decisions to help stem Australia's biodiversity crisis.
"In one collaborative space, the WildObs platform now hosts AI computer vision models which have been trained for Australian animals and environments.
"They can identify hundreds of species in camera trap images, 10 times faster than people.
"In conservation, timing matters and detecting problems early can mean the difference between recovery and extinction."
WildObs uses AI species classifiers to:
- Quickly and cheaply detect rare and elusive species
- Identify if native species are declining earlier
- Assess if invasive species management is effective
- Track biodiversity changes across landscapes and the continent
- Help conservationists prioritise where limited resources should go
Dr Luskin said WildObs was set up to improve national collaboration between scientists, governments and environmental groups working on wildlife monitoring.
"The WildObs platform is an easy end-to-end solution for all researchers," he said.
"Users just upload images and WildObs stores and processes them in the cloud. The results can be downloaded or viewed with interactive dashboards.
"We asked Australian users what they wanted and ecologists worked with an international team of computer scientists to build this platform to suit them."
The image platform was built with QCIF Digital Research, Agouti, Wageningen University, and INBO.
It hosts image classifiers developed by the WildObs-QCIF team along with Google's SpeciesNet, AWC135 from the Australian Wildlife Conservancy, the Tasmanian species recognition model from the University of Tasmania and the Victorian Species Recognition Model by AddaxAI.
"People in Australia were training AI models, but there was no way to easily use them," Dr Luskin said.
"Now anyone can host their AI species classifier on WildObs, allowing new users to access and run it easily, and harness our massive storage and powerful computers.
"Better data use can directly improve conservation outcomes - more effective protection of threatened species, smarter investment in conservation, and stronger environmental reporting."
Collaboration and acknowledgements
WildObs was started with seed money from UQ's Centre for Biodiversity and Conservation Science and School of the Environment . The project is a co-investment partnership between UQ, the Australian Research Data Commons (ARDC), QCIF Digital Research, and the Terrestrial Ecosystem Research Network (TERN). The WildObs image platform was a collaborative project with Agouti, Wageningen University, and INBO in Europe, and we acknowledge their provision of foundational support. WildObs is hosted by the ARDC Nectar Research Cloud. ARDC and TERN are enabled by the Australian Government's National Collaborative Research Infrastructure Strategy (NCRIS). WildObs has been shaped by scientists at universities in all states and territories, national and state governments, and NGOs such as Bush Heritage.
