/ Asahi/Guangzhou, May 23, 2022 -- KubeCon EU 2022 was unveiled as Huawei's open-source cloud-native AI satellite application program
On May 18, at the main forum of KubeCon and CloudNativeCon Europe Summit, a high-profile conference in the cloud native field, Huawei launched the first open source cloud-native AI satellite application solution.
Logic diagram of cloud native AI satellite application scheme based on MindSpore
In recent years, low-earth orbit satellites have become the focus of global space technology. Satellite constellations are formed when many small satellites and cubesats are launched into low earth orbit. The development of global low earth orbit satellites has a broad prospect, but it also faces many technical challenges, such as satellite constellation management, communication management, space traffic management, and opening up new application scenarios such as space mining.
In order to actively cope with the challenges, Beijing University of Posts and Telecommunications and Tianyi Research Institute launched the "Tianyuan Constellation" program, aiming to build an open source in-orbit platform for aerospace computing based on the technology combination of cloud native and artificial intelligence, support the development of satellite technology, and build an integrated digital infrastructure for space and space. The open source cloud native AI satellite application scheme launched at the main forum of KubeCon and CloudNativeCon Europe Summit is based on Huawei's open source KubeEdge cloud native and MindSpore full scene AI framework, which is of pioneering innovative significance.
Using the KubeEdge cloud-native and Rise MindSpore full-scene AI frameworks, Baogeun satellite supports flexible deployment of AI models in terminal, edge, and cloud full-scene scenarios. Baogeun satellite is equipped with orbiting Earth coordination capabilities for new service scenarios such as orbiting Earth coordination image reasoning, incremental deep learning, and joint learning. MindSpore is an open-source full-scene AI framework from Huawei in 2020. It is one of the core software and hardware of the basic AI of Senteng. It supports flexible deployment of full-scene at terminals, edges and cloud sides.
KubeEdge and MindSpore are mainly used to build multi-model collaborative reasoning between ground and satellite, as well as ground incremental model training to facilitate the use of small models on satellites, training in ground data centers and use of global models. When the model encounters difficult samples, the satellite will compress the samples and send them to the ground. In the ground center, the incremental learning task will be carried out using MindSpore to improve the accuracy of the model. The ground center will update and compress the model locally and upload it to the satellite, where Mindspore will decompress and update the model. Enabling satellites to use the latest models for ongoing operations, including forest fire detection and land remote sensing.
In this summit, the open source cloud native AI satellite application solution is the only domestic open source solution selected as the keynote speech of this year's CNCF Europe Summit. MindSpore not only supports the innovative research of scientific basic models such as remote sensing, but also supports cloud native AI satellite application schemes. It is the first open source AI framework in the industry that has the capability of remote sensing AI in the whole scene from satellite to ground artificial intelligence computing center.
In the future, the technology based on cloud-native KubeEdge and Rise MindSpore will also be expanded to more new scenarios in satellite constellations, as well as deep space exploration, such as Earth-orbit, Mars-orbit, Venus-orbit missions, etc.
