Authors: Hasan Nasrallah, Abed Ellatif Samhat, Cristiano Nattero, Ali J. Ghandour

Published on: May 06, 2024

Impact Score: 7.8

Arxiv code: Arxiv:2404.06202

Summary

• What is new: Use of fully convolutional neural networks for multi-class buildings’ instance segmentation from high-resolution satellite images to create national urban maps.
• Why this is important: Lack of proper governance in developing countries to generate and update national rooftop maps affordably and efficiently.
• What the research proposes: An automated pipeline using earth observation and deep learning for producing building maps.
• Results: Successfully created the first comprehensive national building footprint map of Lebanon with approximately 1 million units and 84% accuracy.

Technical Details

Technological frameworks used: Fully convolutional neural networks

Models used: nan

Data used: Sub-meter high-resolution satellite images

Potential Impact

Real estate, urban planning, disaster management, and governance in developing countries.

Want to implement this idea in a business?

We have generated a startup concept here: GeoScan AI.