This study launched the "Where's Whale-do?" machine learning competition in collaboration with NASA, HeroX, Wild Me, and NOAA. Applicants to this competition were asked to apply artificial intelligence to speed and scale photo-based, mark-recapture studies of the Cook Inlet beluga population by building machine learning tools that matched individual beluga whales (Delphinapterus leucas) across photos and years. There are subtle and sometimes changing patterning in individual beluga that can confound automated matching of beluga images, and image quality from variable lighting and water conditions also affected individual beluga identification. Several common techniques were used across competition applicants that exceeded the capabilities of the current state-of-the-art tools and techniques. Ultimately, an ensemble model was created, producing a new matching algorithm named “MiewId” that was deployed it in the Flukebook.org platform for production use. Flukebook.org is now fully operational for Cook Inlet beluga population assessments. In addition to belugas, MiewId has been deployed for identification of a broader array of finned whales and dolphins, which demonstrates a broader benefit from this study.

USD: $182,000

2022-09-01 → 2023-12-31