Case Study: Snow Leopard Trust achieves rapid automated image classification—cutting processing from hours to minutes with Microsoft Azure

Saving snow leopards with deep learning and computer vision on Spark

Microsoft partnered with the Snow Leopard Trust to help study one of the world’s most elusive and endangered cats. Biologists had deployed motion‑sensing cameras across remote Central Asian habitat and collected more than 1 million images, but sorting them by hand was slow, costly and fragmented across multiple spreadsheets—taking around 300 hours per camera survey and diverting resources from conservation work.

Using Azure Machine Learning and Microsoft Machine Learning for Apache Spark, the teams built a scalable image‑classification pipeline that applied transfer learning (ResNet featurization), dataset augmentation, and ensembling across camera bursts. The solution increased accuracy from about 63% to 90%, dramatically reduced false positives, and scaled from hours to a three‑minute run on a 400‑node cluster, freeing hundreds of hours for scientists and delivering a deployable web service for ongoing surveys.

Snow Leopard Trust

Rhetick Sengupta

Board President Snow