一种用于红外目标检测的高效流水线式FPGA加速器  

An ultra-efficient streaming-based FPGA accelerator for infrared target detection

作　　者：陈少毅 汤心溢[2,3,4] 王健 黄静思[1,2,3,4] 李争 CHEN Shao-Yi;TANG Xin-Yi;WANG Jian;HUANG Jing-Si;LI Zheng(School of Information Science and Technology,Shanghai Tech University,Shanghai 201210,China;Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 20083,China;University of Chinese Academy of Sciences,Beijing 100049,China;Key Laboratory of Infrared System Detection and Imaging Technology,Chinese Academy of Sciences,Shanghai 200083,China)

机构地区：[1]上海科技大学,上海201210 [2]中国科学院上海技术物理研究所,上海200083 [3]中国科学院大学,北京100049 [4]中国科学院红外探测与成像技术重点实验室,上海200083

出　　处：《红外与毫米波学报》2022年第5期914-922,共9页Journal of Infrared and Millimeter Waves

基　　金：Supported by the National Pre-Research Foundation of China during the“14th Five-Year Plan”(514010405-207)。

摘　　要：基于深度学习的目标检测算法取得了很大成功,显著超越了传统算法,在很多场景下甚至可以和人类相媲美。不同于可见光相机,红外相机可以在黑暗环境下识别物体,可以用于安防和无人驾驶等领域。本文提出了面向嵌入式设备的轻量级目标检测算法,并采用赛灵思的Ultrascale+MPSoC ZU3EG FPGA加速并部署该算法。加速器运行在350 MHz的时钟频率下,吞吐量达到了551 FPS,功耗仅有8.4 W。在准确率方面,该算法在FLIR数据集下IoU指标达到了73.6%。在性能方面,相比于之前相同逻辑资源下性能最好的硬件加速器Ultranet,该加速器设计将吞吐量提高了2.59倍,功耗降低了2.04倍,降低至原来的49.02%。Object detection algorithm based on deep learning has achieved great success,significantly better than the effect of traditional algorithms,and even surpassed human in many scenarios.Unlike RGB cameras,infrared cameras can see objects even in the dark,which can be used in many fields like surveillance and autonomous driv⁃ing.In this paper,a lightweight target detection algorithm for embedded devices is proposed,which is accelerat⁃ed and deployed using Xilinx Ultrascale+MPSoC FPGA ZU3EG.The accelerator runs at a 350 MHz frequency clock with throughput of 551 FPS and power of only 8.4 W.The intersection over union(IoU)of the algorithm achieves an accuracy of 73.6%on FILR datasets.Comparing with the previous work,the accelerator design im⁃proves performance by 2.59×and reduces 49.02%of the power consumption.

关 键 词：红外图像处理 实时嵌入式系统 可编程逻辑器件 卷积神经网络 

分 类 号：TN47[电子电信—微电子学与固体电子学]