D3325: Invasion Alert: Using machine learning to identify invasive ants

Tuesday, November 12, 2024

9:00 AM – 5:00 PM MST

Location: Phoenix Convention Center, Exhibit Hall 6, PCC

Presenting Author(s)

Lori Lach (she/her/hers)

Professor
James Cook University
Cairns, Queensland, Australia

Co-Author(s)

FZ

Fatima Zaidi

Murdoch University
Murdoch, Western Australia, Australia
HL

Hamid Laga

Murdoch University
Murdoch, Western Australia, Australia
FS

Ferdous Sohel

Murdoch University
Murdoch, Western Australia, Australia
MG

Mahmood Golzarian

Murdoch University
Murdoch, Western Australia, Australia
BH

Ben Hoffmann

CSIRO
Winnellie, Northern Territory, Australia
CB

Chris Burwell

Queensland Museum
South Brisbane, Queensland, Australia
MT

Melissa Thomas

Murdoch University
Murdoch, Western Australia, Australia

Invasive ants are one of the most serious biosecurity risks globally. Early detection and eradication are crucial for limiting the impact of invasive ants, however current practices for identifying invasive ants are time and labour intensive and rely on ever shrinking taxonomic expertise. To help non-specialists determine if an ant is likely to be one seven high priority invasive species, we developed an ant identification platform. The platform uses a hierarchical deep learning-based machine learning approach to identify the seven target ant species or determine that an ant is not a target. The machine learning algorithm has been trained on over 200,000 individual ants from 15,000 images collected from across Australia. Our extensive tests show that the algorithm currently provides 90% or greater correct predictions of the seven target species. Citizen scientists will be able to access this tool using a mobile application on both Android and iOS devices.