Road-DL: Development of a Road Pavement Condition Classifier Utilising Deep Learning Techniques Applied to SAR Data

The objective of ROAD-DL service is to provide road condition assessment based on remote sensing data.  The service aims to provide up to date classification of specific features related with road condition that could trigger inspection or maintenance actions. Specifically, the service ingest Synthetic Aperture Radar (SAR) Sentinel-1 satellite images, provided by ESA through the Copernicus initiative.

Telespazio UK built and tested an approach driven by Deep Learning (DL) to provide road condition assessment to National Highways (NH). The project used multi-temporal SAR Sentinel-1 satellite data and NH collected road condition parameters to train a neural network to identify damaged and degraded roads in the UK strategic road network. The DL classifiers were trained and validated with independent in-situ laser scanner measurement data, which is routinely collected by NH.

The solution is easily scalable and shall act as a continuous monitoring service providing routine insights between ground-based surveys.