Case Study: Snow Leopard Trust achieves faster snow leopard image classification with Microsoft Corporation

Saving snow leopards with deep learning and computer vision on Spark

Snow Leopard Trust, a U.S.-based nonprofit, needed a faster way to sort through more than 1 million motion-camera images used to study endangered snow leopards. Manual classification was taking around 300 hours per camera survey, pulling scientists away from conservation work. The organization worked with Microsoft Corporation using Azure Machine Learning, Microsoft Machine Learning for Apache Spark, and Microsoft Cognitive Toolkit.

Microsoft Corporation built a scalable image classification workflow using deep learning, transfer learning, dataset augmentation, and sequence-based prediction across Spark. The solution cut image processing time from hours to about three minutes on a 400-node cluster, reached 90% accuracy, nearly eliminated false positives, and saved hundreds of hours of manual work so biologists could focus more on conservation.

Snow Leopard Trust

Rhetick Sengupta

Board President Snow