Satellite imaging swarms with distributed onboard intelligence

Descripción del proyecto

Sistemas de imagen e inferencia por satélite de nueva generación

La unión hace la fuerza en los sistemas de vigilancia espaciados. Múltiples satélites con inteligencia distribuida pueden recoger imágenes simultáneas desde distintas ubicaciones o ángulos, que luego pueden combinarse para proporcionar representaciones completas de escenas tridimensionales. Los cálculos de inteligencia artificial de estas representaciones pueden realizarse a bordo de los satélites y transmitirse a los enlaces de comunicación terrestres casi en tiempo real. En este contexto, el proyecto IntelliSwarm, financiado por el Consejo Europeo de Investigación, pretende desarrollar modalidades de cooperación que faciliten la inferencia a bordo distribuida por satélite y mejoren la calidad de las imágenes. Ello implica el desarrollo de varias técnicas innovadoras, como la mejora de la resolución espacial y espectral, así como el aumento tridimensional, la compresión basada en la inteligencia artificial (IA) y las metodologías de inferencia distribuida para la IA, así como modelos de IA rápidos y energéticamente eficientes para alimentar la informática distribuida a bordo.

Objetivo

Space-based monitoring systems for Earth observation are being increasingly employed by industry and governments, and large constellations having tens to hundreds of units will be deployed in the near future. This project considers swarms of satellites in formation flight, and it investigates innovative cooperation modalities to enhance image quality and enable satellite distributed onboard inference. Through artificial intelligence (AI), multiple simultaneous and redundant satellite image acquisitions from different viewing angles will be cooperatively combined into scene representations having significantly higher spatial and spectral resolution and augmented with surface profile information, enabling imaging beyond the diffraction limit by using very high-resolution training sets. These representations will be computed onboard the satellites in a distributed fashion and will be used for inference and image reconstruction; they will allow to increase quality, availability and timeliness of the data and to optimize the usage of inter-satellite and satellite-to-ground communication links. Enabling this cooperation requires major scientific advances; the project will generate new knowledge via innovative scene models allowing spatial/spectral resolution enhancement and 3D augmentation, innovative AI-based compression techniques, novel distributed inference methodologies for AI, as well as a new generation of fast and energy-efficient AI models powering distributed onboard computing. This ambitious research will become the cornerstone of next-generation satellite imaging and inference systems, increasing scientific and economic value for the users and allowing public and private entities to exploit advanced image products and analytics that can benefit the economy and the society. The developed methods will also be relevant to several other types of embedded systems, further extending the project impact to areas such as sensor networks and Internet of Things.