基于深度学习的SKA图像反卷积研究

Study on SKA Image Deconvolution Using Deep Learning

干涉测量使得观测天文图像的分辨率显著提升,然而其阵型所带来的点扩展函数效应需要图像反卷积技术来消除。尽管传统CLEAN反卷积算法已经广泛应用于射电天文图像的点扩展函数消除,但仍然存在精度不高的问题。为了解决国际大科学工程——平方公里阵(SKA)的图像模糊问题,本文提出一种深度卷积神经网络来提升射电天文图像重建的质量。实验显示,相较于通用的方法,本文提出的方法能更好地重建弱源,并在整体图像质量上有明显提升。

Radio interferometry makes the observing resolution of astronomical images significantly improved, but deconvolution is required to eliminate the effects of the point spread function (PSF). Although the traditional CLEAN-based deconvolution has been widely used to eliminate the observed PSF, it still has the problem of low accurate reconstruction. To solve the PSF problem of the square kilometer array (SKA), a deep convolutional neural network is proposed to improve the quality of radio image reconstruction. Experiments show that compared with the traditional method—CLEAN, the method proposed in this paper can better reconstruct weak sources and significantly improve the quality of an image.