Satellite image generation through Synthetic Aperture Radar or SAR technology allows for high spatial resolution due to the relative motion of the sensor that simulates a long antenna.

This imaging process has several advantages over optical systems because it enables the generation of high-resolution images even at night or in cloud-covered areas. Due to these peculiarities, SAR plays a key role in remote sensing and planet monitoring operations.

«With this project we worked on developing an innovative system, capable of simulating such computer images, to train Automatic Target Recognition or ATR software based on deep learning» explains Achille Peternier, Adjunct Professor at the Institute of Information Systems and Networking (ISIN). «This is particularly useful, for example, to create surveillance units to identify ships or carry out search and rescue operations.»

The SAR image simulator is based on a transposition into the radar signal of the "ray tracing" technique commonly used in 3D computer graphics to obtain highly realistic (optical) images.

Through the use of graphics accelerators (GPUs), realistic SAR-type images can be generated quickly and inexpensively, making it possible to effectively meet the need for large data collections for training ATR systems.

«The procedural creation of datasets for artificial intelligence-based systems provides an immediate answer to the growing problem of collecting the data needed for their training, while also ensuring an economically sustainable and legally and ethically compliant approach» Peternier continues.

«This simulator represents the state of the art currently available on the market and gives us a significant competitive advantage, especially when we consider the increasingly present and important role that artificial intelligence plays in all sectors» Giaime Origgi, project manager sarmap SA.