American satellite imagery operator Planet Labs and graphics processor manufacturer NVIDIA have announced a strategic alliance. The companies plan to equip spacecraft with powerful computing modules capable of processing images without the participation of ground centers. This will allow you to receive ready-made analytical data in a matter of seconds instead of the usual hours.
Every day, the Planet satellite constellation scans the entire surface of the Earth, accumulating petabytes of information. Traditional CPU-based processing can no longer cope with this volume. The use of NVIDIA graphics accelerators with CUDA architecture will speed up key operations – orthorectification, decluttering and composite frame assembly.
Computing power will be distributed between cloud servers, ground stations and directly on-board systems. The satellites integrate the NVIDIA Blackwell and IGX Thor platforms, which minimizes the delay between shooting and issuing the result. This is critical for monitoring emergencies and fast-moving events.
Image quality will also reach a new level. NVIDIA CorrDiff diffusion technology based on generative neural networks, will provide a super-resolution effect. The model, trained on years of Planet archives, uses physical algorithms to restore the smallest details even in archival images.
The third area of cooperation is the creation of global embeddings. These are vector representations of multidimensional data that will turn a continuous stream of satellite observations into something like a digital map of the world. The system will gain the ability to perform semantic search on a planetary scale, automatically identifying complex patterns and anomalies faster than any existing algorithms.
Of particular interest is testing the NVIDIA IGX Jetson Thor module for space applications. These chips are planned to be installed on the new generation of Pelican satellites and in the promising Owl constellation. Transferring calculations to orbit will make it possible to send ready-made results to Earth instead of raw data sets, which will further speed up the receipt of insights.
The cooperation is developing against the backdrop of a tightening of Planet Labs’ policy in the field of image distribution. In March, the company imposed a two-week delay or a complete ban on publishing images of the Middle East. The official reason is the risk of revealing the movements of the US military in the context of the conflict with Iran. The decision limited previous policies of openness but underscored the growing value of space imaging in geopolitical struggles.
Integration artificial intelligence in space imaging is radically changing the approach to security and analytics. Satellites are no longer just eyes in orbit, turning into full-fledged computing centers.
Source: new-science.ru
Image: new-science.ru
