“Hey CSIRO, what bug is this?” It’s a question that scientists at Australia’s national science agency often hear.
And, in a country home to 170 species of snakes and 2000 kinds of spider, it’s easier asked than answered.
So where do you go when you want the details on what’s creeping and crawling near you?
A CSIRO collaboration with Data61 is seeing the construction of a ‘Critterpedia’ tool that uses machine learning and citizen science to take the guesswork out of animal spotting.
The tool allows users to take photos of a snake or spider from their smart device. A system trained with an algorithm then classifies it, providing information on the family, genus or species.
The artificially intelligent (AI) platform considers not only these images, but also additional information, like GPS location.
To teach the platform, CSIRO says hundreds of thousands of images of snakes and spiders were fed into the system – a sizeable task “uniquely perfect for an AI solution”.
“The visual differences between the two species can be quite subtle. We need a great deal of training data to adequately identify critters,” explains project lead and Data61 researcher, Dr Matt Adcock.
“We’ve started off with an enormous amount of images sourced from zoological experts collaborating with Critterpedia, and have developed a suite of tools to help semi-automatically label these images, verify the information, and cross check with other data sources.”
He said the application aims to provide education and awareness for all Australians.
“And, as a wildlife safety tool, could ultimately save human and animal lives.”
You can currently sign up to become a Phase 1 tester, which lets you download a beta version of the platform and submit wildlife photos to keep training the algorithm.
Click through to read a full version of this article on the CSIRO’s Algorithm blog.
