阙钰佳,
周泽南,
周远远,
张晓玲,
孙铭芳
电子科技大学信息与通信学院 ??成都 ??611731
基金项目:国家自然科学基金(61671113)
详细信息
作者简介:师君:师 君(1979–),男,河南南阳人,博士,电子科技大学信息与通信工程学院副教授,主要从事SAR成像技术、雷达信号处理研究,已发表论文50余篇。E-mail: shijun@uestc.edu.cn
阙钰佳(1993–),男,福建龙岩人,电子科技大学硕士生,主要从事阵列3维SAR成像及目标识别技术研究。E-mail: queyujia1993@163.com
周泽南(1997–),男,海南海口人,电子科技大学硕士生,主要从事深度学习技术在SAR图像的应用研究。E-mail: 2942714332@qq.com
周远远(1992–),男,山东金乡人,电子科技大学博士生,主要从事深度学习技术在SAR图像的应用研究。E-mail: 732156543@qq.com
张晓玲(1964–),女,四川成都人,博士,电子科技大学信息与通信工程学院教授,主要从事SAR成像技术、雷达探测技术研究,已发表论文50余篇。E-mail: xlzhang@uestc.edu.cn
孙铭芳(1981–),男,辽宁本溪人,硕士,北京华航无线电测量研究所高工,从事无线电导航与引信,SAR成像,雷达信号处理研究。E-mail: sunmf125@163.com
通讯作者:师君 shijun@uestc.edu.cn
责任主编:陈杰 Corresponding Editor: CHEN Jie中图分类号:TN957.52
计量
文章访问数:1728
HTML全文浏览量:823
PDF下载量:328
被引次数:0
出版历程
收稿日期:2018-10-22
修回日期:2019-07-02
网络出版日期:2019-08-23
Near-field Millimeter Wave 3D Imaging and Object Detection Method
SHI Jun,,QUE Yujia,
ZHOU Zenan,
ZHOU Yuanyuan,
ZHANG Xiaoling,
SUN Mingfang
School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China
Funds:The National Natural Science Foundation of China (61671113)
More Information
Corresponding author:SHI Jun, shijun@uestc.edu.cn
摘要
摘要:主动式毫米波阵列3维成像系统是人体安检成像系统的研究热点,该文对主动式毫米波阵列3维系统工作模式、信号模型和成像算法进行了介绍,并将深度学习中的卷积神经网络(CNN)热图检测方法和边框回归检测技术应用于人体安检成像异物检测。研究表明,基于热图的检测方法和基于YOLO的检测方法均可实现异物检测。基于热图的检测方法网络结构简单、易训练,但由于需要遍历整幅待检测图像,运算时间长,且生成的检测框尺寸固定,无法适应异物尺寸变化。基于YOLO的检测算法网络结构复杂、训练耗时长,但该方法在检测速度与检测框精度上优势明显,更利于机场安检等对实时性要求较高的检测应用。
关键词:近场毫米波3维成像/
后向投影/
卷积神经网络/
图像检测/
边框回归
Abstract:Active mm-wave linear-array 3D imaging system has become one of the active research areas in the field of imaging for human security. In this paper, the operating mode, signal model, and imaging algorithm are introduced. Deep learning algorithms, including the Convolutional Neural Network (CNN) with heat map and You Only Look Once (YOLO) network, were used for the object detection of human security image. The results show that the method based on heat map and YOLO can both detect foreign objects. We find that the CNN with heat map has a simple network construction and can be easily trained, but the detection process needs to traverse the whole image, which is relatively time-consuming, and the size of the detection region cannot adapt to the objects. On the contrary, though with a relatively complex construction, YOLO network has advantages in terms of detection efficiency and accuracy. Furthermore, the size of the detection region can adapt to the objects, which is more suitable for the human security imaging application.
Key words:Near-filed millimeter wave 3D imaging/
Back projection/
Convolutional Neural Network (CNN)/
Objects detection/
Boxing regression
PDF全文下载地址:
https://plugin.sowise.cn/viewpdf/198_e452513f-b49f-49ae-9839-918b104c857d_R18089