Wildlife Insights sharpens its vision with updated AI model and helps treasured Australian wildlife recover

Google Earth
Dec 15, 2020 · 3 min read

By Tanya Birch, Program Manager, Google Earth Outreach

Today, we are announcing new updates to Wildlife Insights that includes new features and enhancements to the AI model that powers the Wildlife Insights platform, one of the initiatives of Google Earth Outreach’s Nature Conservation program. Wildlife Insights is a collaboration between Google, Conservation International, and 6 leading conservation organizations, to provide a cloud-based software platform to manage, identify and share wildlife images taken from motion-triggered cameras, also called camera traps, which are placed in areas with lots of wildlife.

With our mission to save biologists’ time analyzing species data and deriving more meaningful insights on biodiversity data, we hope these features will make camera trap data a little more manageable for biodiversity scientists.

These new model features include expanding the number of species included to 732. We have moved to a new model architecture called EfficientNet which increases the accuracy of predictions by 2% across all species on average. Starting today, we also filter our predictions geographically so we won’t predict an African elephant if your camera trap is located in Asia.

The image on the left shows Leopardus weidii, or a Margay as the common name, instead of “No CV Result” due to the upgrade to EfficientNet model architecture.

We also added a feature that shows a prediction higher up the taxonomic levels, instead of just at species level, when we aren’t confident about the species. This reduces the amount of time we don’t show any predictions by up to 40%.

The image on the left shows a higher taxonomic level above “species”, an improvement over when the model could not return a classification for higher taxonomic levels (image on the right).

Learn more about our species identification models on the Wildlife Insights About our AI site.

More wildlife images are added to our Explore map every day — the largest public collection of wildlife “selfies” in the world. Anyone can explore candid camera photos of wildlife in the most remote regions of our planet, including Australia, where Wildlife Insights is being used as part of a $1M Google.org grant to measure and analyze whether and where wildlife is returning due to World Wild Fund for Nature (WWF) Australia’s post-wildfire recovery efforts.

Wildlife Insights is a partnership between Conservation International, Smithsonian Institution, North Carolina Museum of Natural Sciences, Wildlife Conservation Society, WWF, Zoological Society of London and Yale University’s Map of Life Lab, with Google as founding technology partner. Wildlife Insights is supported by the Gordon and Betty Moore Foundation, the Lyda Hill Foundation, Google, the World Bank and the McGovern Foundation.