Input: high-resolution satellite imagery (roughly 0.3–0.5 m/pixel) · Output: boxes or points classed as large or small vehicle.
We detected and classified vehicles from high-resolution satellite imagery using a deliberately simple, robust split: large vehicles (trucks, buses, heavy equipment) and small vehicles (cars, motorcycles, small vans).
A hard problem at small scale
Satellite-based detection already contends with tiny objects, occlusions and variable orientation. Adding a size-based split asks the model to separate visually similar roof shapes by scale and context alone — a subtle distinction at this resolution.
How it works
A single-stage detector (in the YOLO family) trained on imagery annotated with just two classes automatically locates each vehicle and assigns its size class. Even at satellite scale, that turns one frame into traffic and activity intelligence across very large areas — no ground sensors required.
