摘要
以提升毫米波全息图像分辨率为目的,提出基于深度学习网络的毫米波全息成像图像重建方法。设计毫米波全息成像系统,通过信号划分与混频滤波放大处理等过程计算目标对像回波信号的相位和幅度,将获取的数据传输至全息图像成像程序,得到毫米波全息图像。利用基于深度学习网络的图像重建算法对毫米波全息图像实施重建:处理毫米波全息图像数据,构建外部图像库;构建深度学习网络中的卷积神经网络模型,通过训练获取低分辨率图像与高分辨率图像间的映射关系;基于输入卷积神经网络模型的毫米波全息图像低分辨率图像块,获取高分辨率毫米波全息图像。采用残差结构优化卷积神经网络模型解决梯度弥散现象。测试结果显示该方法重建图像分辨率的两个评价参数图像重建精度与重建图像清晰度分别为97.5%和97.1%。
In order to improve the resolution of millimeter wave holographic image,a method of millimeter wave holographic image reconstruction based on deep learning network is proposed.A millimeter wave holographic imaging system is designed.The phase and amplitude of the target image echo signal are calculated through signal division,mixing filtering and amplification processing.The obtained data are transmitted to the holographic image imaging program to obtain the millimeter wave holographic image.The millimeter wave hologram is reconstructed by image reconstruction algorithm based on deep learning network:Processing millimeter wave hologram data and constructing external image library;constructing convolution neural network model in deep learning network to obtain the mapping relationship between low resolution image and high resolution image through training;millimeter wave hologram image based on input convolution neural network model is low High resolution image block is used to obtain high resolution millimeter wave hologram.The residual structure is used to optimize the convolution neural network model to solve the gradient dispersion phenomenon.The test results show that the two evaluation parameters of reconstructed image resolution are 97.5%and 97.1%respectively.
作者
罗梦贞
俞春强
LUO Mengzhen;YU Chunqiang(Lijiang College,Guangxi Normal University,Guilin Guangxi,541006;School of Computer Science and Information Engineering,Guangxi Normal University,Guilin Guangxi,541004)
出处
《激光杂志》
CAS
北大核心
2021年第6期68-72,共5页
Laser Journal
基金
广西高校中青年教师科研基础能力提升项目(No.KY2016LX556)。
关键词
深度学习
毫米波
全息成像
图像重建
分辨率
卷积神经网络
deep learning
millimeter wave
holographic imaging
image reconstruction
resolution
convolution neural network
