Uncharted Technology
Guiding Machine Learning to Classify Satellite Imagery

Locust infestations have a devastating effect on crop yields. Multispectral imaging from orbital satellites can reveal environmental conditions that create optimal environments for locust breeding, but the inherently disjointed data hinders traditional AI and ML approaches to discovering them.

In this application of Distil, we developed a workflow that—given just a small selection of human-labelled images—guides self-supervised machine learning to build extensive training datasets and several candidate locust breeding ground models. In evaluations with agencies tasked with assessing infestations of crop land, biologists lacking data science experience were easily able to increase model accuracy and generate actionable insights from satellite data in a way that was previously unattainable.