Project
Global Change Detection from Satellite Imagery (IARPA SMART)
Automating the analysis of Landsat, Sentinel, and WorldView imagery to detect and characterize large-scale change anywhere on Earth.
The IARPA SMART program (Space-based Machine Automated Recognition Technique) set an ambitious goal: automatically detect, characterize, and monitor anthropogenic change — construction, land conversion, infrastructure growth — across the entire planet, using the combined stream of Landsat, Sentinel, and commercial WorldView imagery.
AICIP joined a national team coordinated by Accenture Federal Services, contributing the lab’s remote-sensing machine learning expertise: harmonizing observations across sensors with wildly different resolutions and revisit rates, and building models that recognize change processes rather than just pixel differences. Five lab members worked on the effort across its span.
The program’s fingerprints are visible in the lab’s subsequent research: the multi-scale representation challenges SMART surfaced — the same site imaged at 30m and 0.3m must mean the same thing to a model — directly motivated the lab’s Cross-Scale MAE work on scale-robust self-supervised pre-training for overhead imagery.