MAS project: Active Satellite Monitoring

MAS (Spanish for Active Satellite Monitoring) is an artificial intelligence (AI) solution that uses convolutional neural networks and computer vision to monitor changes in critical infrastructure.

The use case of MAS consists of the surveillance and continuous monitoring of points or areas of interest using remote sensing Earth observation images, typically from satellites. It includes the analysis of those images through AI tools and georeferenced image processing techniques.

The images are provided by the Copernicus project of the European Space Agency, which allows free and periodic access to most of the planet’s emerged surface. They are delivered by a server belonging to an internal tool developed for another project called REPOSAT, consisting of making satellite images available. REPOSAT was subsidized by RED.ES as part of the Subsidies program for the Development of Technological Offer for Digital Content 2020, with file number 2020/0820/00102245.

Thanks to the way in which the images have been obtained, we have been able to develop a solution without any additional cost to access the data, which limits the economic effort to the computational resources of the development and maintenance of the tool.

We have developed two different AI models: one based on convolutional neural networks for monitoring known and well-identified anomalies, and another on regulatory analysis using unsupervised learning for wide-spectrum detection of unidentified anomalies. This allows adaptation to possible future changes in the characteristics of the structures to be monitored. The changes would happen through transfer learning with the convolutional network model, and due to its own design characteristics with the unsupervised vision model.