An Australian environmental group has received federal government funding to train AI models how to accurately recognise and identify up to 120 native species.
The Australian Wildlife Conservancy (AWC) announced Tuesday it had received A$750,000 ($489,000) from Canberra’s A$8 million Innovative Biodiversity Monitoring Grants Programme.
The group said it would use the funds to create open-source AI recognisers that will assist in effectively processing camera trap data collected during biodiversity surveys across the country.
“AWC is increasingly turning to emerging technologies to improve efficiency and the quality of data collected in the field,” AWC CEO Tim Allard said.
“By harnessing the power of AI, these species recognisers will provide an accurate and cost-effective method for monitoring changes in biodiversity across Australia. It will also help free our ecologists to spend more time in the field, where they have the biggest impact.”
While a single camera trap can collect thousands of images, it takes days or even months for ecologists to manually review each image bank, AWC noted.
AI has huge potential to improve the speed and cost effectiveness of camera trap image processing, the group said.
Challenges still remain, however, due to the number of high-quality target species images which are required across a range of habitats to accurately train the AI model.
AWC said its national science team had started gathering images from its extensive back catalogue that ranges back over the last 15 years from sanctuaries and project partners, in preparation for the AI project.
The group’s data scientists and software developers have also developed a camera trap processing and reporting pipeline and tested methods by successfully creating AI recognisers for 44 species to date.
With the federal grant funding secured, AWC will develop species recognisers for 120 mammals and reptiles, including the endangered Northern Bettong, the near-threatened Western Quoll, the vulnerable Great Desert Skink, and the Yellow-spotted Monitor.
The species classifier models will be made available on a public open-source repository, such as GitHub, along with documentation on how to use the model for animal classification, as well as documentation on AWC’s camera trap image processing pipeline, methods and tools.
“This project represents excellent value for government investment as it builds on AWC’s existing sanctuary and data collection infrastructure, bolstered by an established network of expert ecologists familiar with the different fauna assemblages of each region,” Allard said.
“These resources cannot be replicated without millions of dollars of investment and years of effective project planning.”
AWC was one of 12 universities and NGOs that were awarded a parcel of the A$8 mln from the government’s programme.
Other recipients included the World Wildlife Fund for Nature Australia, Australia National University, the CSIRO, and the Indigenous Desert Alliance.
MAKING PREPARATIONS
According to the government, the grants would support the establishment of its Nature Repair Market, and support the integrity of market outcomes by developing effective ways to monitor biodiversity.
The projects awarded grant funding would provide “innovative biodiversity baseline assessment and monitoring technologies and approaches for use in the NRM”, according to the Department of Climate Change, Energy, Environment and Water’s website.
Since the government successfully legislated the overarching framework for the NRM, it has been working with parties to establish expert reference groups and committees to oversee the scheme, as well as consulting on subordinate legislation and legislative rules.
Additionally, it has been developing what it described as a biodiversity assessment instrument and methods, including convening workshops with external partners.
The government has previously indicated it hopes to see the market operational as soon as this year.
By Mark Tilly – mark@carbon-pulse.com
** Click here to sign up to our twice-weekly biodiversity newsletter **
