Using remote cameras, the Snow Leopard Trust has captured more than 1.3 million images and ongoing and planned camera trapping efforts will add another 500,000 images each year.
Since the cameras are triggered by any type of movement, they often take images of goats, sheep, birds, blowing leaves of grass, etc. Only about 5 percent of the pictures actually contain a snow leopard, and often it’s hard to recognize a shape for what it is. Manually reviewing each digital file to find the images that contain a snow leopard can take thousands of hours of staff time.
To solve this resource challenge, the Trust and Microsoft have employed deep learning, a technique that uses what engineers have called “brainlike algorithms” that learn – and then teach themselves – to sort through the photos automatically.
Using these algorithms, Microsoft engineers have created a scalable image recognition program that is already nearly 95 percent accurate in identifying snow leopards in camera trap photos.
The team additionally created a live dashboard that highlights snow leopard hot spots where the cats particularly like visiting and leaving marks.
“This is a huge step forward for us, but the work does not end here”, says Rhetick Sengupta, president of the Snow Leopard Trust’s board of directors and a principal program manager at Microsoft. “The next challenge will be for the image recognition program to learn to identify and distinguish between individual snow leopards based on their fur patterns, which is the basis for coming population estimates and monitoring. Currently, this is done manually, by researchers and volunteers. Automating this second step of the process will not only save an enormous amount of time and resources, but will also make estimates more reliable.”
Before the AI for Earth grant, Microsoft engineers had been working on the project pro bono. The $15,000 grant will now allow both partners to expand the work.
The snow leopards’ habitat spans 12 countries and covers an area that’s more than twice the size of the state of Texas. According to latest estimates, there are between 3,900-6,500 snow leopards in the wild.
However, these numbers are fraught with uncertainty. “It’s an undisputed fact that no one knows what the numbers are,” Rhetick Sengupta told GeekWire in a recent interview.
This uncertainty is not just an academic problem. There is a real danger that decisions about the animals’ protection are being made based on weak data. The type of tools Microsoft and the Snow Leopard Trust are developing under this AI for Earth grant could help change that.
AI for Earth is a Microsoft program aimed at empowering people and organizations to solve global environmental challenges by increasing access to AI tools and educational opportunities, while accelerating innovation. AI for Earth provides Azure computing resources and AI tools to NGOs, academic institutions and research organizations that work on challenges in climate change, biodiversity, agriculture and water.
The Snow Leopard Trust is the world’s oldest and largest organization dedicated to the conservation of the endangered snow leopard. It joins a growing number of AI for Earth grantees worldwide recognized by Microsoft for the impact and potential.