The present situation and developing trends of space-based intelligent computing technology
-
摘要:
太空任务的蓬勃发展，极大地促进了航天应用对人工智能技术的需求.智能算法在航天器上的应用、助推航天器的智能化能力的提升，已成为目前航天领域的发展趋势.但目前航天器的智能算法计算能力仍然不足，严重制约了航天」领域的智能化发展.针对典型的航天智能应用，分析了智能应用对计算能◣力的需求，并调研了航天智能计算领域的研究现状，针对目前智能计算技术对航天智能应用的支撑情况进行了归纳总结，在此基础上指出了航天领域中智能计算技术发展应走“计算芯片系列化、计算〗平台通用化、配套软件统一化”的道路，构建高能效的智能计算平◇台和完整丰富的航天智能基础生态.本文所提出的天基智能计算技术领域中待解决的关键技术及技术战略发展路线，对航天领域在智能化变革之际抓住机遇、推动新一轮航天产业革命有着重要的意义.
Abstract:The progress of space missions has promoted the development of artificial intelligence technology in the aerospace field. Specifically, the deployment of intelligent algorithms on spacecraft to enhance the intelligence capabilities of spacecraft has become the current trend. However, at present, spacecraft lacks sufficient computing power for intelligent algorithms, which severely restricts the development of the aerospace intelligence. This article analyzes the computing power demand of typical aerospace intelligent applications, and introduces the current research status in the field of aerospace intelligent computing. Moreover, the work summarizes the current intelligent computing technological state for aerospace intelligent applications. On this basis, the paper points out that the future development of aerospace intelligent computing technology should take the path of serialization of computing chips, universalization of computing platform and unification of supporting software to create a product form which is equipped with intelligent computing platforms and complete aerospace intelligent basic ecology. The key technical issues to be solved and the development strategy proposed in this paper are of great significance for seizing the opportunity to promote a new round of space industry revolution.
-
Key words:
- spacecraft /
- artificial intelligence /
- space-based /
- aerospace application /
- intelligent computing
-
图 4 天基智能计算技术发展展望
Figure 4. Development prospect of space-based intelligent computing technology
表 1 不同应用场景对计算量的需求统计
Table 1. Calculation demand statistics for different application scenarios
航天智能应用 采用的♂神经
网络算法运行速度要求
(单位为时间、帧率)输入图像尺寸
(长×宽)计算能力∮需求：
GMCC/s内存需求
(中间▂结果存储要求)外存需求
(模型大小)视觉深度估计 super point < 500ms 640×512 75.16 217.09MB 5.48MB ResNet18 FCN 航天器型号
及事件识别跟踪
(孪生神经网】络)25FPS 640×512 153.5 17.2MB 403.14MB 100B 400B LSTM 4×4 623.36KB 664.7KB 超分辨率重建 10FPS 16×16 2.16 701.59KB 3.21MB mobilenet 船检测 YOLO v1 25FPS 300×300 239.75 13.87MB 171.04MB SSD 25FPS 101 134.04MB 5.8MB 云检测 Mask RCNN 1s 1K×1K 11.05 396.76MB 24.18MB MobileNet 1s 8K×8K 697.04 25.28GB 24.18MB 遥感图像
目标检测Mask RCNN 1s 1K×1K 11.05 396.76MB 24.18MB Mobile Net 1s 8K×8K 697.04 25.28GB 24.18MB 
下载: 导出CSV
表 2 国内外典型智能计算平台的性能比对表
Table 2. Performance comparison table of domestic and international typical intelligent computing platforms
公司 产品 AI算力 功耗 尺寸 芯片规格 工作温度 内存 Aitech S-A1760
Venus1 TFLOPS，能效
比：60 GFLOPS/W8W~20W 127×129×52mm Pascal GPU+双核
NVIDIA Denver 2@
2.0GHz+4核ARM
A57 @ 2.0GHz-40~65℃ 8GB
LRDDR4
(128 bit)英特尔 Movidius 13.9GMACC/s 1W 72.5×27×14mm Movidius Myriad
VPU0~40℃ 4GB LPDDR3
(32bit)百度 EdgeBoard
FZ93.6TOPS 12W~30W 220×126.5×50mm Zynq UltraScale+
ZU9EG0~50℃ 8GB
DDR4华为 Atlas 200 22/16/8 TOPS 4GB: 5.5 W
8GB: 8W52.6×38.5× 8.5mm 8核ARM
A55@1.6GHz+2个
DaVinci AI核-25~80℃ 8 GB/4 GB
LPDDR4X寒武纪 思元220-M.2 8TOPS(INT8) 8.25W 80 × 22 × 7.2mm 双核ARM
A55+MLUv02 AI核-20~80℃ LPDDR4x
64 bit鲲云科技 雨人AI
加速卡102.4 GOPS 7.0~8.5W 核心板：50 × 60mm 双核ARMA9 +
CAISA Engine0~70℃ 1GB DDR4
下载: 导出CSV
-
[1] 张海金, 苏若皓, 崔媛媛, 等. 面向军用电子系统的高性能高可靠处理器设计技术[J]. 微电子学与计算机, 2020, 37(3): 66-70. DOI: 10.19304/j.cnki.issn1000-7180.2020.03.013.ZHANG H J, SU R H, CUI Y Y, et al. High-Performance and reliable processor design technology for military electronic system[J]. Microelectronics & Computer, 2020, 37(3): 66-70. DOI: 10.19304/j.cnki.issn1000-7180.2020.03.013. [2] 黎藜, 乔德治, 何世民, 等. 星载计算机软件通用测试平台的设计与实现[J]. 微电子学与计算机, 2019, 36(3): 23-27. DOI: 10.19304/j.cnki.issn1000-7180.2019.03.005.LI L, QIAO D Z, HE S M, et al. Design and implementation of universal software test platform for onboard computer[J]. Microelectronics & Computer, 2019, 36(3): 23-27. DOI: 10.19304/j.cnki.issn1000-7180.2019.03.005. [3] 刘兴潭, 武延鹏. 天基空间目标识别方法综述[C]//第二届中国空天安全会议论文集. 大连: 中国指挥与控制学会空天安全平行系统专业委员会, 2017: 42-50.LIU X T, WU Y P. A survey of space-based objects identification algorithms[C]//The 2nd China Aerospace Safety Conference. Dalian: Professional Committee of Aerospace Safety parallel system of China Command and Control Society, 2017: 42-50. [4] 李众, 白瑞君, 洪军, 等. 基于自适应特征融合及多尺度输出的遥↓感图像飞机检测算法[J]. 微电子学与计算机, 2021, 38(4): 40-45. DOI: 10.19304/j.cnki.issn1000-7180.2021.04.008.LI Z, BAI R J, HONG J, et al. Aircraft detection algorithm of remote sensing image based on adaptive feature fusion and multi-scale output[J]. Microelectronics & Computer, 2021, 38(4): 40-45. DOI: 10.19304/j.cnki.issn1000-7180.2021.04.008. [5] ACCESSPATH. Planet公司利用卫星和人工智能〖进行地理空间情报搜集[EB/OL]. (2019-01-23)[2021-09-23]. .ACCESSPATH. Planet uses satellites and artificial intelligence for geospatial intelligence gathering[EB/OL]. (2019-01-23)[2021-09-23]. . [6] NASA. The enhanced thematic mapper plus(ETM+)[EB/OL]. [2021-09-17]. . [7] ZHAO PY, LIU JG, WU CC. Survey on research and developmentof on-orbit active debris removal methods[J]. Science ChinaTechnological Sciences, 2020, 63(11): 2188-2210. [8] SHILOV A. Nvidia'S Jetson AI board is ready to go to space[EB/OL]. (2021-08-14)[2021-09-23]. . [9] AITECH. S-A1760 VenusTM | radiation-characterized space AI GPGPU[EB/OL]. (2021-08-14)[2021-09-26]. . [10] INTEL. Intel MovidiusTM MyriadTM 2 vision Processing unit 4GB[EB/OL]. [2021-09-26]. . [11] XLINX. Radiation tolerant kintex UltraScale XQRKU060 FPGA data sheet[EB/OL]. (2020-12-17)[2021-10-12]. . [12] 李丽婷. 人工智能芯片技术进展及产业发展研究报告[J]. 厦门科技, 2019(1): 1-9. DOI: 10.3969/j.issn.1007-1563.2019.01.001.LI L T. A survey of artificial intelligence ch-iptechnology progress and industrial develop-ment[J]. Xiamen Science & Technology, 2019(1): 1-9. DOI: 10.3969/j.issn.1007-1563.2019.01.001. [13] XLINX. Zynq UltraScale+ MPSoC product tables and product selection guide[EB/OL]. [2021-09-17]. . [14] AI市场. EdgeBoard嵌入式AI计算卡(FZ9)[EB/OL]. [2021-09-22]. .Baidu. Edgeboard embedded AI computingcard (FZ9)[EB/OL]. [2021-09-22]. . [15] 寒武纪. 思元220边缘计算♀模组①[EB/OL]. (2019-11-14)[2021-10-14]. .CAMBRICON. Siyuan 220-M. 2 edge end AI accelerator card[EB/OL]. (2019-11-14)[2021-10-14]. . [16] HUAWEI. Atlas 200 AI加速模块[EB/OL]. (2021-06-01)[2021-10-14]. .HUAWEI. Atlas 200 AI accelerator[EB/OL]. (2021-06-01)[2021-10-14]. . [17] 鲲云科技. 鲲云科技CAISA芯片入选世界人工智能大会最高奖SAIL奖TOP30[EB/OL]. (2021-06-11)[2021-10-14]. .CORERAIN. CAISA chip was selected for the TOP30 SAIL award of the world artificial intelligence conference[EB/OL]. (2021-06-11)[2021-10-14]. . [18] 鲲云科技. 雨人加速卡[EB/OL]. [2021-09-17]. .CORERAIN. Rainman accelerator[EB/OL]. [2021-09-17]. . [19] 王朴, 周晴. 基于龙芯1E的嵌入式Linux实时性的优化与可靠性设计[J]. 微电子学与计算机, 2019, 36(11): 11-15. DOI: 10.19304/j.cnki.issn1000-7180.2019.11.003.WANG P, ZHOU Q. Real-time optimization and reliability Realization of embedded Linux based on Loonson 1E[J]. Microelectronics & Computer, 2019, 36(11): 11-15. DOI: 10.19304/j.cnki.issn1000-7180.2019.11.003. [20] 复旦微电子集团. 高可靠集成电路产品及应用解决方案[EB/OL]. [2021-09-29]. .Fudan Microelectronics Group Company Limited. Highly reliable integratedcircuit products and application solutions[EB/OL]. [2021-09-29]. . -
下载:
点击查看大图
计量
- 文章访问数: 389
- HTML全文浏览量: 99
- PDF下载量: 64
- 被引次数: 0
