부산대학교 도서관

Based on accelerating the application of artificial intelligence technology in the aerospace field, this paper takes remote sensing satellite on-orbit perception and recognition, planning decision-making and autonomous execution requirements as traction, and analyzes that the on-board intelligent computing platform should include CPU, DSP, NPU, FPGA and public storage resources. On the basis of studying the status of traditional on-board computing platforms, a three-bus switching network hardware architecture with both high reliability and high performance is proposed. Aiming at the efficient utilization of multi-type resources in the computing platform, a software platform for unified scheduling of heterogeneous resources is proposed. Building an intelligent computing platform from the two aspects meets future intelligent applications and can be developed for users easily. In the long run, it provides a research foundation for the intelligence of future spacecraft.