Our recent projects in the lab have increasingly included native bees and describing how their diversity and assemblages vary as a function of the management of farmland and the landscape around farms. These studies depend in large part on the ability to identify native bees to species. For anyone who has tried this identification of insects to species can be quite tricky and time consuming. There are likely over 500 species of native bees just here in Wisconsin and many of them look to the untrained eye pretty much identical.
Frustrated by the slow pace of getting bees identified by specialists – there are just a handful of experts here in North America who are capable of reliably identifying our bees – a couple of years ago, I (Claudio) approached some colleagues in Electrical and Computer Engineering and told them of our challenges and asked the naive question, “There must be some automated way of identifying our bees, no?” After all, we can identify tanks from spy satellites in space or people’s faces in our photo cataloging software, how hard can a bee be? Well, as it turns out, this isn’t that easy either!
My colleague Prof. Bill Sethares in ECE thought this could be a interesting project for his Image classification and Image processing classes. To make a long a story short, several students, including MS student Chris Hall, thought this was interesting enough and eventually develop a series of computer algorithms to first extract features from bee wings based on venation patterns and then classify the bees into groups, i.e., species. Chris completed his Master’s degree on this project last fall and is now at Sandia National Labs. For more on this project and where we hope to take it check our their cool video they put together explaining their project:
If you want to learn more go over to here. I hope we will have more on this topic in later years as we continue our collaboration.
Posted by Hannah Gaines and Claudio Gratton