基于视觉处理器芯片的遥感图像智能处理系统设计与验证  

Design and Implementation of Remote Sensing Image Processing System with Intelligent Vision Processor

作　　者：于双铭 武文波[3] 窦润江 刘力源[1,4] 刘剑 吴南健[1,2] YU Shuangming;WU Wenbo;DOU Runjiang;LIU Liyuan;LIU Jian;WU Nanjian(State Key Laboratory of Superlattices and Microstructures,Institute of Semiconductors,Chinese Academy of Sciences,Beijing 100083,China;University of Chinese Academy of Sciences,College of Materials Science and Opto-Electronic Technology,Beijing 100049,China;Beijing Institute of Space Mechanics&Electricity,Beijing 100094,China;University of Chinese Academy of Sciences,School of Electronic Electrical and Communication Engineering,Beijing 100049,China)

机构地区：[1]中国科学院半导体研究所半导体超晶格国家重点实验室,北京100083 [2]中国科学院大学材料科学与光电技术学院,北京100049 [3]北京空间机电研究所,北京100094 [4]中国科学院大学电子电气与通信工程学院,北京100049

出　　处：《航天返回与遥感》2023年第5期54-64,共11页Spacecraft Recovery & Remote Sensing

基　　金：国家自然科学基金(62274154,U20A20205,U21A20504);国家重点研发计划(2019YFB2204300);中国科学院青年创新促进会项目(2021109)。

摘　　要：随着遥感卫星成像技术向高分辨率、高帧率的方向快速发展,星上遥感图像处理技术面临海量数据处理速度慢、有效信息提取能力不足、载荷功耗高等瓶颈问题。文章提出一种面向星上处理的遥感图像智能处理系统,系统架构设计基于一款边缘型视觉处理器芯片,该芯片为可重构并行处理器架构,可通过指令编程兼容支持图像预处理算法、计算机视觉算法和深度学习神经网络算法。基于算法-硬件协同设计思想,对图像处理算法进行了剪枝和量化等压缩优化,在芯片上实现算法的高效部署。搭建基于视觉处理器芯片的遥感图像智能处理系统,进行实验验证。结果表明,该系统能够完成对遥感图像的实时目标检测等智能化处理,系统处理能力达到200 M像素/s,芯片峰值计算能效达到2×1012次/W(300 MHz时)。With the rapid development of remote sensing satellite imaging technology,image processing technology is faced with challenges,such as slow data processing,detection and high power consumption.This paper presents a remote sensing image processing system with a vision processor chip for intelligently processing remote sensing images.The chip has a dynamically reconfigurable parallel architecture is proposed to flexibly and efficiently implement image preprocessing,computer vision algorithms and deep learning neural network algorithms.The algorithm/hardware co-design method is used to compact algorithm models by pruning and quantization techniques.The processing system for remote sensing image processing is built and tested with real remote images.The test results show that the vision processor designed and implemented in the paper achieves an energy efficiency of 2TOPS/W@300 MHz clock frequency.The system can realize remote sensing image target detection function with a performance of 200M pixel/second.

关 键 词：遥感图像 图像处理 视觉处理器 深度学习神经网络 

分 类 号：V445[航空宇航科学与技术—飞行器设计]