IRIX4US: Super-Resolution Algorithm IRIX4 Applied to Urban Settlements

The perimeter that delimits a building or vertical structure in urban areas is called a Building Footprint (BF). BFs are often produced by photointerpretation and digitalisation of remote sensing images and are often used to feed advanced Artificial Intelligence models. Nowadays, BFs are automatically extracted from remotely sensed images using Deep Learning (DL) algorithms. These algorithms depend on the spatial resolution of the input image, because that the higher it is, the higher accuracy values the algorithm can reach. However, more often than not, these images are not available frequently and at regular intervals, even when provided by commercial operators.

The satellites available in the Copernicus Program, especially the Sentinel-2 constellation, provide free and open images with a high temporal resolution, at a 5 day-revisit interval. This is potentially a valuable resource that allows city governments and stakeholders to study the urban dynamics. However, these images usually do not have the spatial detail needed for BF extraction.

The present project aims to perform a wall-to-wall study of urban dynamics by developing a pipeline integrating three state-of-the-art technologies:

• Implementing a validated Deep Learning model to downscale (super-resolve) Sentinel-2 images 4 times, named “IRIX4 Super Resolution algorithm”, preserving spectral characteristics while enhancing the spatial resolution.
• Perform an exhaustive BF extraction from super-resolved Sentinel-2 images, based on advanced DL techniques and segmentation procedures.
• Automate change detection processes over selected urban areas between the years 2020 and 2023 with the use of ad-hoc models.

Pilot areas and final scope of algorithms will be set according to the needs of key stakeholders at different scales and data governance.