From Competition to Collaboration: Automated Identification of Right Whales
Christin Khan, Jason Holmberg, Philip Hamilton, Heather Pettis, Drew Blount, Jason Parham, Marcin Mucha, Claire Charlton, Victoria Rowntree, Els Vermeulen, Will Rayment, Steve Dawson, David Johnston, Karina Groch
International Whaling Commission
Artificial intelligence (AI) software for photo ID matching of right whales has progressed through multi-national collaboration with NOAA, Kaggle, Deepsense.ai, Wild Me and right whale researchers. AI algorithm training was undertaken using 53,266 aerial images of North Atlantic Right Whales from the New England Aquarium database. The algorithm of choice Deepsense.ai is now achieving 88.9% top-1 accuracy and 97.8% top-5 accuracy using high quality aerial imagery. The photo identification software is operational online in the Flukebook platform. Further algorithm training is currently underway using southern right whale images from South Africa, Argentina, Brazil, Australia, and New Zealand. The software is available via the online Flukebook platform for public use for photo identification matching of North Atlantic right whales now and will be operational for southern right whales by the end of 2020 to complete phase 1. Phase 2 includes algorithm training using lateral shots and this will be undertaken during Q4 of 2020.