GAN-Based Focusing-Enhancement Method for Monochromatic Synthetic Aperture Imaging

被引：5

作者：

Ye, Guoyao ^{[1
]}

Zhang, Zixin ^{[1
]}

Ding, Li ^{[1
,2
,3
]}

Li, Yinwei ^{[2
,4
]}

Zhu, Yiming ^{[1
,2
,3
]}

机构：

[1] Univ Shanghai Sci & Technol, Terahertz Technol Innovat Res Inst, Shanghai 200093, Peoples R China

[2] Univ Shanghai Sci & Technol, Shanghai Key Lab Modern Opt Syst, Shanghai 200093, Peoples R China

[3] Terahertz Sci Cooperat Innovat Ctr, Shanghai 200093, Peoples R China

[4] Tongji Univ, Shanghai Inst Intelligent Sci & Technol, Shanghai 200092, Peoples R China

来源：

IEEE SENSORS JOURNAL | 2020年 / 20卷 / 19期

基金：

中国国家自然科学基金;

关键词：

Imaging; Generators; Sensors; Apertures; Generative adversarial networks; Gallium nitride; Standards; MMW near field imaging; monochromatic full-focus; SAR; image fusion; GAN-FEM;

D O I：

10.1109/JSEN.2020.2996656

中图分类号：

TM [电工技术]; TN [电子技术、通信技术];

学科分类号：

0808 ; 0809 ;

摘要：

Two-dimensional (2-D) synthetic aperture imaging with a single frequency suffers from limited depth-of-focus (DOF), and leads to the difficulty of focusing volume targets. In this paper,as opposed to using a wide band for 3-D imaging, this out-of-focus problem is examined as a multi-focal imaging issue. To solve the limited DOF problem, we propose a generative adversarial network (GAN) based focusing-enhancement method (GAN-FEM) to fit an unknown out-of-focus kernel for MMW monochromatic synthetic aperture imaging. To determine which type of MMW-images dataset of input can be better suitable for GAN, the grayscale and pseudo-color images dataset are tested respectively to train the neural network. Proof-of-principle experiments are performed at 94 GHz and the results prove that our proposed GAN-FEM can greatly improve the focusing performance for volume targets. The effectiveness of our proposed method confirms the focusing-enhancement capacity of 2-D monochromatic imaging system for 3-D targets, and provides a possible solution to reduce the system complexity for practical 3-D imaging missions.

引用

页码：11484 / 11489

页数：6