Our team developed and implemented the technology of recognizing satellite images based on Artificial Intelligence (AI). 

Building shapes are created using an automated production process of object recognition from high-resolution multispectral satellite images. Our Machine Learning methods, developed based on Convolutional Neural Networks and Deep Learning Techniques, allow 3D and 2D shapes extracting for the entire country fast and at a high accuracy level.

PRODUCT FEATURES:

  • 95% of buildings with an area > 25 sq.m matched automatically through the ML algorithm
  • 100% coverage completeness is achieved due to manual post-processing and data validation
  • X, Y, Z accuracy: ± 3m

Depending on the project’s needs, LoD 1.0/1.1, LoD 1.2/1.3 or LoD2 building footprints could be produced representing buildings and roof details with multiple heights.

The building footprints layer is delivered separately or integrated into the final product (3D city map, Digital Surface Model, Digital Height Model, Clutter Heights map, City population map)