MHI Develops an Onboard AI-Based Object Detector Utilizing a Next-Generation Space-Grade MPU
In-Orbit Demonstration to be Performed aboard Innovative Satellite Technology Demonstration-4
- Onboard AI-Based Object Detector comprises of an AI-equipped data processor and an earth observation camera developed by the Tokyo University of Science - Demonstration of object detection by in-orbit AI processing and remote AI updating from the ground
TOKYO, Mar 6, 2024 - (JCN Newswire) - Mitsubishi Heavy Industries, Ltd. (MHI) recently developed an onboard artificial intelligence (AI)-based object detector called "AIRIS"(1) that detects objects from satellite images with satellite-mounted AI driven by the next-generation space-grade MPU(2) "SOISOC4." AIRIS is scheduled to be launched aboard the small demonstration satellite "RAISE-4"(3) in FY2025, after which it will conduct an in-orbit demonstration.
AIRIS is a satellite-mounted device consisting of an AI-equipped data processor and an earth observation camera developed by the Tokyo University of Science. It detects objects by taking images of objects on the earth's surface from orbit and then processing the images with AI. By using AI in-orbit to perform detection processing-which heretofore has been performed on the ground-AIRIS can use its AI to select and transmit to the ground only those areas where target objects are visible in images. It can also receive an AI model that was "retrained" on the ground based on the transmitted images and update its own onboard AI while in orbit.
AIRIS will execute a demonstration utilizing these features as part of the "Innovative Satellite Technology Demonstration-4"(4) mission, which is scheduled for launch in FY2025. This mission is being executed as a demonstration opportunity under the "Innovative Satellite Technology Demonstration Program" by the Japan Aerospace Exploration Agency (JAXA). Specifically, AIRIS will execute a cycle of taking images of vessels sailing on the surface, using AI to select only areas in the images that contain vessels, transmitting those areas to the ground, having the AI undergo retraining on the ground, and then updating the AI by sending the retrained AI model to the AI in orbit.
AIRIS operations will be controlled with a next-generation space-grade MPU "SOISOC4" that is the product of development undertaken by JAXA and MHI. SOISOC4 is a domestically produced microprocessor that was designed and manufactured based on advanced SOI (Silicon on Insulator) and SOC (System on Chip) design technologies for general consumer use. It has high radiation resistance to withstand the harsh radiation environment of deep space, various communication functions, and security functions, and can operate with low power consumption.
MHI will conduct an in-orbit demonstration of SOISOC4 together with an AI operation demonstration of AIRIS as part of the in-orbit demonstrations planned for the "Innovative Satellite Technology Demonstration-4" mission. Through its development of these devices, MHI will contribute to Japan's space development and utilization efforts by further enhancing the technologies and expertise it has accumulated to date in developing cutting-edge space equipment.
(1) Stands for "Artificial Intelligence Retraining In Space." (2) Abbreviation of "Micro Processing Unit." (3) Stands for "RApid Innovative payload demonstration SatellitE-4." RAISE-4 is one of the satellites that will be launched aboard JAXA's Innovative Satellite Technology Demonstration-4 mission. (4) AIRIS was selected as the subject of demonstration for Innovative Satellite Technology Demonstration-4 in June 2023.
Mitsubishi Heavy Industries (MHI) Group is one of the world’s leading industrial groups, spanning energy, smart infrastructure, industrial machinery, aerospace and defense. MHI Group combines cutting-edge technology with deep experience to deliver innovative, integrated solutions that help to realize a carbon neutral world, improve the quality of life and ensure a safer world. For more information, please visit www.mhi.com or follow our insights and stories on spectra.mhi.com.