基于FPGA和CNN的水下目标识别系统  

Underwater target recognition system based on FPGA and CNN

作　　者：陈凯峰 梁鉴如[1] 陈强[1] 傅伟清 刘贵庆 刘露露 CHEN Kaifeng;LIANG Jianru;CHEN Qiang;FU Weiqing;LIU Guiqing;LIU Lulu(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学电子电气工程学院,上海201620

出　　处：《传感器与微系统》2021年第4期103-105,109,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(61272097);上海市科技委员会重点项目(18511101600);上海工程技术大学研究生科研创新资助项目(E3-0903-18-01051)。

摘　　要：水下智能设备由于功率受限,采集的图像数据无法进行实时处理。考虑现场可编程门阵列(FPGA)功耗低,计算能力强和灵活性高等特点,基于FPGA的并行性和流水线技术,利用卷积神经网络(CNN)强大的图像处理能力,设计了一个低功耗图像实时识别系统。实验使用三个不同水域的水下视频图像对该系统进行验证。实验结果表明:该系统达到了与PC工作站相同的精度,实现1920×1080分辨率和25FPS帧速率图像的处理,满足水下识别任务的需求。Due to the limited power of underwater intelligent device,acquired image data cannot be processed in real time.Consider characteristics of low power consumption,high computing power and high flexibility of field programmable gate array(FPGA).FPGA-based parallelism and pipeline technology,use powerful image processing capability of convolutional neural network(CNN),a low power consumption image real-time recognition system is designed.The system is verified uses underwater video images from three different water regions.Experimental results show that the system achieves the same precision as the PC workstation,and realize processing of 1920×1080 resolution and 25FPS frame rate image to meet the needs of underwater recognition tasks.

关 键 词：卷积神经网络 现场可编程门阵列 图像识别 

分 类 号：TN710[电子电信—电路与系统]