Using AI to push the boundaries of wildlife survey technologies

wildlife-monitoringsatellite-imagerydeep-learningbiodiversityremote-sensing

Abstraction: Deep learning counts 500,000 wildebeest from satellite imagery automatically

Key points:

Tiejun Wang and Zijing Wu (University of Twente, ITC Faculty) built a deep learning model to automatically locate and count large herds of migratory ungulates in the Serengeti-Mara ecosystem
Used fine-resolution satellite imagery (38–50 cm) to detect nearly 500,000 individual wildebeest and zebra across thousands of square kilometers and multiple habitat types
Largest training dataset ever published for a satellite-based wildlife survey: 53,906 annotated individuals
Method is open-source, spatially scalable, and transferable to other open-landscape wildlife surveys
First demonstrated capability for total counts of migratory ungulates from space across highly heterogeneous landscapes
Published in Nature Communications (2023); DOI: 10.1038/s41467-023-38901-y

Connections: University Of Twente · Deep Learning · Remote Sensing · Computer Vision