Kawasaki and Yamaguchi, Japan, Nov 27, 2025 – (JCN Newswire via SeaPRwire.com) – Fujitsu Limited and Yamaguchi University today announced the successful development of a low-power edge computing technology that enables image processing on small satellites using redundant GPUs within 10 minutes (near real time). This innovative technology is designed for low-Earth orbit Synthetic Aperture Radar (SAR) satellites, which illuminate the Earth’s surface with microwaves and receive reflected waves to create two-dimensional images.
The new system offers high fault tolerance against cosmic radiation and operates efficiently within the typical 20W power constraints of small satellites. It can process data, including error detection and reprocessing, within approximately 10 minutes. Validated with raw SAR data on a satellite prototype, the technology successfully performed both L1 [1] (raw data to 2D image) and L2 [2] (image to observed quantity) processing, accurately determining ocean surface wind speeds. This versatile technology is also applicable to optical and multi-hyperspectral satellites.
Fujitsu plans to release the developed programming environment, Fujitsu Research Soft Error Radiation Armor, to users in Japan in February 2026.
Moving forward, Fujitsu and Yamaguchi University will continue to refine correction data accuracy. Fujitsu aims to advance services and operations by enabling near real-time AI processing on satellites, realizing previously unfeasible applications. This will involve in-orbit validation on actual satellites to promote user-friendly data processing systems across various satellite types.
About the new technology
Fujitsu and Yamaguchi University leveraged Fujitsu’s computing technologies, cultivated through its expertise in AI and supercomputer development, and Yamaguchi University’s satellite data (remote sensing data) analysis technologies to develop the following key functions:
1. Highly fault-tolerant computer system with redundant configuration operating within satellite power limits
- System employs redundant processors to detect errors, a critical feature for operating within the challenging space environment for efficient
- To meet satellite power constraints, the system dynamically manages computing resources and program execution, ensuring both performance and energy efficiency
2. Development of a programming environment for computer systems resilient to cosmic radiation errors
- Robust programming environment (library) using Linux, Python, and open-source software
- Library simplifies the implementation of error detection, computer restarts, and recalculation functions, and includes a novel technique to improve error processing efficiency by dividing computational jobs
Figure 1: Difference in processing time of two errors.
For simplicity, restart time is not included.
A prototype system, operating under 20W, successfully processed raw SAR data (L1 and L2) in under 10 minutes, achieving near real-time performance. By applying compression, correction, and wind speed estimation models to the time distribution of radar reflection intensity from the ground, ocean surface wind speeds could be calculated in units of several hundred meters (see Figure 2 below). Real-time calculation of ocean surface wind speeds from satellites is expected to contribute to maritime safety by providing immediate notifications of high-wind areas to vessels.
Figure 2: Raw data (left), L1 processed data (center), and L2 processed data of sea surface wind speed (right). For the L2 processed data, only the sea area is displayed, with wind speed represented by color. Ships and bridges are also converted as windy areas and would typically be removed as noise [3]
Background
Satellites, orbiting in the range of 200 km – 36,000 km from the Earth’s surface, offer extensive Earth observation using UV, IR, and radar for diverse applications like object detection, environmental monitoring, and resource mapping. However, traditional ground-based data processing introduces multi-hour delays. Small satellites face strict power limits (typically under 20W) and require robust error detection against cosmic radiation, which conventional programming struggles to address.
[1] L1 Processing:
Process to calculate the Earth’s surface conditions from the time distribution of radar reflection intensity, also known as compression processing.
[2] L2 Processing:
Process to apply corrections based on information about the Earth’s surface and atmosphere to L1 processed data, which then derives physical quantities such as ocean surface wind speed and wave height from the state of the reflected waves.
[3] These figures contain modified Copernicus Sentinel data [2025].
About Fujitsu
Fujitsu’s purpose is to make the world more sustainable by building trust in society through innovation. As the digital transformation partner of choice for customers around the globe, our 113,000 employees work to resolve some of the greatest challenges facing humanity. Our range of services and solutions draw on five key technologies: AI, Computing, Networks, Data & Security, and Converging Technologies, which we bring together to deliver sustainability transformation. Fujitsu Limited (TSE:6702) reported consolidated revenues of 3.6 trillion yen (US$23 billion) for the fiscal year ended March 31, 2025 and remains the top digital services company in Japan by market share. Find out more: global.fujitsu
Press Contacts
Fujitsu Limited
Inquiries
Yamaguchi University
(For inquiries regarding research results)
Organization for
Research Initiatives, Yamaguchi University
Prof. Dr. NAGAI Masahiko
E-mail nagaim*yamaguchi-u.ac.jp (replace * with @)
(For media inquiries)
Public Relations, Yamaguchi University
E-mail sh011*yamaguchi-u.ac.jp (replace * with @)
Copyright 2025 JCN Newswire via SeaPRwire.com. All rights reserved. www.jcnnewswire.com