Automatic dragonfly identification

Species identification underlies biological studies, informs management and conservation, and connects people with the natural world around them, but it’s not always easy. New, game-changing smartphone “apps” like LeafSnap and Merlin Bird ID have made this task easier than ever, and the machine learning systems that power them have the potential to revolutionize the way we interact with life on Earth, but no system like this yet exists for identifying dragonflies and damselflies. Here, we present Odomatic, a system that accurately classifies Odonata from images of their wings, and the Targeted Odonata Wing Digitization (TOWD) initiative aimed at digitizing the wings of North American odonate taxa, as well as some Central and South American species. Odomatic, trained on the TOWD dataset, will be able to identify most of the Odonata of the New World, and the dataset will provide a unique, extensive resource for studying wing evolution in Odonata. Finally, we will introduce another exciting extension of Odomatic, currently under development, allowing it to identify Odonata from real-world photographs by leveraging the vast store of images in OdonataCentral.org. Funding provided by NSF grants 1611642 and 1564386.