12.04.24 - GeoUrban-AI: An Automated Solution for Efficient Urban Mapping
Discover GeoUrban-AI tool from GEOAI, a web-based tool which autonomously extracts building footprints from satellite and aerial imagery.
What is the real-world application of the GeoUrban-AI tool?
It's a powerful resource for developing countries that frequently lack the means to generate and update national rooftop maps. Traditional methods are often costly and time-consuming, but this tool is proposed as a solution, bridging the gap with an automated pipeline for national urban maps.
This approach utilizes earth observation and deep learning methods to achieve high object-wise accuracy for buildings' instance segmentation from high-resolution satellite images. The GeoUrban-AI tool showcases three CNN-based deep learning models users can interact with:
- General Model: Trained on a satellite dataset (WHU) with a 30 cm-resolution using their multi-class instance segmentation technique
- Sci-Net Model: Trained on a drone dataset with a spatial resolution of 2 cm to 20 cm, using their scale-invariant semantic segmentation technique
- Lebanese Model: Trained on their Lebanese satellite dataset with a 50 cm spatial resolution, using the multi-class instance segmentation technique
The team deployed their application in Lebanon, where the first comprehensive national building footprint map was produced, covering approximately 1 million units with 84% accuracy.
Take a look at this tool and the possibilities it opens up for urban planning and mapping : https://lnkd.in/eYgfRUxw
Hashtag#GeoUrbanAI Hashtag#GeospatialAnalysis Hashtag#UrbanPlanning Hashtag#DeepLearning