基于残差条件生成对抗网络的SKA图像重建  

SKA Image Reconstruction Based on Residual Conditional Generative Adversarial Networks

作　　者：陈丽绵 张利 王蓓 蔡宇佳 许丽 CHEN Li-mian;ZHANG Li;WANG Bei;CAI Yu-jia;XU Li(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025

出　　处：《软件导刊》2023年第3期162-167,共6页Software Guide

基　　金：国家重点研发计划项目(2022YFE0133700,2018YFA0404602,2018YFA0404603);国家自然科学基金项目(11963003,12273007);SKA专项资助项目(2020SKA0110300);贵州省科技厅项目(黔科合基础-ZK[2022]一般143);贵州大学培育项目(贵大培育[2020]76号);贵州大学引进人才科研基金项目(贵大人基合字(2018)60号)。

摘　　要：射电干涉阵列的天线数目有限,不能在傅里叶空间中获取完整的天空图像信息,使得通过对测量数据进行傅里叶反变换得到的重构图像中出现伪影,造成射电源的细节模糊不清。为进一步提升图像重建质量,采用一种基于残差条件生成对抗网络的射电图像复原方案,并将其应用于SKA模拟数据。模型通过优化训练后能够从欠采样下的退化图像中恢复原始图像。实验结果表明,与常用的射电图像洁化算法相比,该深度学习算法在图像指标PSNR和SSIM上均有显著提升,能够更好地恢复天体的亮度分布图。Due to the limited number of antennas in the radio interference array,the complete sky image information can not be obtained in the Fourier space,which makes the reconstructed image obtained by inverse Fourier transform of the measured data appear artifacts,resulting in the ambiguity of the details of the radio source.In order to further improve the quality of image reconstruction,adopt a radio image restoration scheme based on residual condition generation adversarial network and applies it to SKA simulation data.The model can recover the original image from the degraded image under undersampling after optimization training.Experimental results show that compared with the commonly used radio image clean algorithm,the proposed deep learning algorithm has a significant improvement in the image index PSNR and SSIM,and can better restore the brightness distribution map of celestial bodies.

关 键 词：干涉成像 图像重建 SKA图像复原 射电观测 条件生成对抗网络 残差网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]