一种星载图像智能处理装置设计与实现  

Design and Implementation of a Satellite-Borne Image Intelligent Processing Device

作　　者：伍攀峰 吴宝林[2] 王允森 朱其星 王明贺 郭清源 杨宁 徐明道 WU Panfeng;WU Baolin;WANG Yunsen;ZHU Qixing;WANG Minghe;GUO Qingyuan;YANG Ning;XU Mingdao(Shandong Institute of Space Electronic Technology,Yantai 264000,China;Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]山东航天电子技术研究所,烟台264003 [2]哈尔滨工业大学,哈尔滨150001

出　　处：《空间控制技术与应用》2022年第5期78-85,共8页Aerospace Control and Application

基　　金：装发重点实验室基金项目(61425040205)。

摘　　要：为提升卫星在轨智能处理能力与国产化率,采用国产智能芯片为核心构建星载图像处理装置硬件平台,并在该平台上设计、部署了一种基于YOLOV3的改进型深度学习算法,同时,为解决国产智能芯片的空间可靠应用问题,进行了抗辐照加固设计,并采用热管和散热板相结合的方式进行散热.该装置图像处理速度相比NVIDIA公司的Jetson TX2 GPU处理速度提升2~4倍,6类典型目标识别率均优于85%,可实现卫星光学图像目标在轨实时识别,通过热仿真分析与试验验证表明,该装置设计可满足在轨使用要求.In order to improve the in-orbit intelligent processing capability and localization rate of satellites,a hardware platform for onboard image processing device is constructed via domestic intelligent chips as the core,and an improved deep learning algorithm based on YOLOV3 is designed and deployed on the platform.At the same time,in order to solve the problem of reliable application of domestic intelligent chip in space,the anti-radiation reinforcement design is carried out,and the heat pipe and heat dissipation plate are combined to carry out heat dissipation.The image processing speed of this device is 2~4 times faster than that of NVIDIA Jetson TX2 GPU,and the recognition rate of six typical targets is better than 85%,which could realize the real-time recognition of satellite optical image targets in orbit.Thermal simulation analysis and experimental verification show that the device can meet the requirements of in-orbit use.

关 键 词：星载 图像 智能处理 国产智能芯片 

分 类 号：V443.5[航空宇航科学与技术—飞行器设计]