Case Study: Whale Watch Azores achieves more accurate sperm whale identification with Capgemini

Using AI to protect whales

Whale Watch Azores needed a better way to identify and track endangered sperm whales from fluke photos, since researchers were manually cataloging more than 2,200 individuals and the process was slow, noisy, and difficult. In the Capgemini Global Data Science Challenge, Capgemini used ML and image recognition to tackle this whale-identification problem with a solution that could help scientists monitor individual whales more efficiently.

Capgemini built an ML pipeline using exploratory data analysis, image cropping, and a combination of classification and Siamese networks with a ResNet-101 backbone. The result was strong performance, with 88% top-1 accuracy, 91% top-3, and 94% top-20, plus a 3% lead over the second-place team on the private leaderboard. Capgemini then productionized the system on AWS so researchers could upload images and get matches in about 2.5 minutes from a cold start, making whale identification much faster and more practical.

Whale Watch Azores

Lisa Steiner

Marine Biologist