Observations of astrophysical objects such as galaxies are limited by various
sources of random and systematic noise from the sky background, the optical
system of the telescope and the detector used to record the data. Conventional
deconvolution techniques are limited in their ability to recover features in
imaging data by the Shannon-Nyquist sampling theorem. Here we train a
generative adversarial network (GAN) on a sample of $4,550$ images of nearby
galaxies at $0.01<z<0.02$ from the Sloan Digital Sky Survey and conduct
$10\times$ cross validation to evaluate the results. We present a method using
a GAN trained on galaxy images that can recover features from artificially
degraded images with worse seeing and higher noise than the original with a
performance which far exceeds simple deconvolution. The ability to better
recover detailed features such as galaxy morphology from low-signal-to-noise
and low angular resolution imaging data significantly increases our ability to
study existing data sets of astrophysical objects as well as future
observations with observatories such as the Large Synoptic Sky Telescope (LSST)
and the Hubble and James Webb space telescopes.

使用生成对抗网络训练的方法，能够更好地恢复低信噪比和低角分辨率成像数据中的天体物体的详细特征，从而增加研究天体物体的现有数据集以及未来观测中的能力。

生成对抗网络恢复超越去卷积极限的星系天体图像特征

Generative Adversarial Networks recover features in astrophysical images  of galaxies beyond the deconvolution limit

With the advent of infrared long-baseline interferometers with more than two
telescopes, both the size and the completeness of interferometric data sets
have significantly increased, allowing images based on models with no a priori
assumptions to be reconstructed. Our main objective is to analyze the multiple
parameters of the image reconstruction process with particular attention to the
regularization term and the study of their behavior in different situations.
The secondary goal is to derive practical rules for the users. Using the
Multi-aperture image Reconstruction Algorithm (MiRA), we performed multiple
systematic tests, analyzing 11 regularization terms commonly used. The tests
are made on different astrophysical objects, different (u,v) plane coverages
and several signal-to-noise ratios to determine the minimal configuration
needed to reconstruct an image. We establish a methodology and we introduce the
mean-square errors (MSE) to discuss the results. From the ~24000 simulations
performed for the benchmarking of image reconstruction with MiRA, we are able
to classify the different regularizations in the context of the observations.
We find typical values of the regularization weight. A minimal (u,v) coverage
is required to reconstruct an acceptable image, whereas no limits are found for
the studied values of the signal-to-noise ratio. We also show that
super-resolution can be achieved with increasing performance with the (u,v)
coverage filling. Using image reconstruction with a sufficient (u,v) coverage
is shown to be reliable. The choice of the main parameters of the
reconstruction is tightly constrained. We recommend that efforts to develop
interferometric infrastructures should first concentrate on the number of
telescopes to combine, and secondly on improving the accuracy and sensitivity
of the arrays.

该研究利用多孔径成像重构算法进行多项测试，分析图像重构的多个参数，并讨论它们在不同情况下的行为，同时推导出实用规则，结果表明，通过使用足够的 u，v 覆盖范围，图像重构可靠，并建议在干涉仪架构的开发中关注望远镜数量的组合，以及提高阵列的精度和灵敏度。