Deep learning for a space-variant deconvolution in galaxy surveys

Sureau, F.; Lechat, Alexis; Starck, Jean-Luc

doi:10.1051/0004-6361/201937039

Cited by 34 publications

(41 citation statements)

References 59 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Starck et al (2002) and Puetter et al (2005) provide thorough reviews of deconvolution methods commonly used in astronomy, which are numerous. New methods are still being actively developed for applications in astronomy (see, e.g., Sureau et al 2020). In comparison to some algorithms, a major advantage of this method is that it works on a per-exposure basis.…”

Section: Comparison To Similar Methodsmentioning

confidence: 99%

Superresolution Reconstruction of Severely Undersampled Point-spread Functions Using Point-source Stacking and Deconvolution

Symons

Zemcov

Bock

et al. 2021

ApJS

View full text Add to dashboard Cite

Point-spread function (PSF) estimation in spatially undersampled images is challenging because large pixels average fine-scale spatial information. This is problematic when fine-resolution details are necessary, as in optimal photometry where knowledge of the illumination pattern beyond the native spatial resolution of the image may be required. Here, we introduce a method of PSF reconstruction where point sources are artificially sampled beyond the native resolution of an image and combined together via stacking to return a finely sampled estimate of the PSF. This estimate is then deconvolved from the pixel-gridding function to return a superresolution kernel that can be used for optimally weighted photometry. We benchmark against the <1% photometric error requirement of the upcoming SPHEREx mission to assess performance in a concrete example. We find that standard methods like Richardson-Lucy deconvolution are not sufficient to achieve this stringent requirement. We investigate a more advanced method with significant heritage in image analysis called iterative back-projection (IBP) and demonstrate it using idealized Gaussian cases and simulated SPHEREx images. In testing this method on real images recorded by the LORRI instrument on New Horizons, we are able to identify systematic pointing drift. Our IBP-derived PSF kernels allow photometric accuracy significantly better than the requirement in individual SPHEREx exposures. This PSF reconstruction method is broadly applicable to a variety of problems and combines computationally simple techniques in a way that is robust to complicating factors such as severe undersampling, spatially complex PSFs, noise, crowded fields, or limited source numbers.

show abstract

Section: Comparison To Similar Methodsmentioning

confidence: 99%

Superresolution Reconstruction of Severely Undersampled Point-spread Functions Using Point-source Stacking and Deconvolution

Symons

Zemcov

Bock

et al. 2021

ApJS

View full text Add to dashboard Cite

show abstract

“…Lanusse et al 2016;Peel et al 2017) and radio interferometric data (e.g Pratley et al 2018), blind source separation of optical and radio sources (e.g. Joseph et al 2016;Jiang et al 2017), and recently in combination with deep learning (e.g Sureau et al 2020).…”

Section: Sparsity and Starlet Regularisationmentioning

confidence: 99%

“…Deep learning techniques would likely accelerate the algorithm by replacing steps considered as bottlenecks (see e.g. Meinhardt et al 2017;Sureau et al 2020). Second, by using other non-linear solutions to explore the lens model parameter space.…”

Section: Lens Model Optimisationmentioning

confidence: 99%

SLITRONOMY: Towards a fully wavelet-based strong lensing inversion technique

Galan

Peel

Joseph

et al. 2021

A&A

Self Cite

View full text Add to dashboard Cite

Strong gravitational lensing provides a wealth of astrophysical information on the baryonic and dark matter content of galaxies. It also serves as a valuable cosmological probe by allowing us to measure the Hubble constant independently of other methods. These applications all require the difficult task of inverting the lens equation and simultaneously reconstructing the mass profile of the lens along with the original light profile of the unlensed source. As there is no reason for either the lens or the source to be simple, we need methods that both invert the lens equation with a large number of degrees of freedom and also enforce a well-controlled regularisation that avoids the appearance of spurious structures. This can be beautifully accomplished by representing signals in wavelet space. Building on the Sparse Lens Inversion Technique (SLIT), we present an improved sparsity-based method that describes lensed sources using wavelets and optimises over the parameters given an analytical lens mass profile. We applied our technique on simulated HST and E-ELT data, as well as on real HST images of lenses from the Sloan Lens ACS sample, assuming a lens model. We show that wavelets allowed us to reconstruct lensed sources containing detailed substructures when using both present-day data and very high-resolution images expected from future thirty-metre-class telescopes. In the latter case, wavelets moreover provide a much more tractable solution in terms of quality and computation time compared to using a source model that combines smooth analytical profiles and shapelets. Requiring very little human interaction, our flexible pixel-based technique fits into the ongoing effort to devise automated modelling schemes. It can be incorporated in the standard workflow of sampling analytical lens model parameters while modelling the source on a pixelated grid. The method, which we call SLITRONOMY, is freely available as a new plug-in to the modelling software LENSTRONOMY.

show abstract

“…Moreover, deep learning deconvolution is used in astronomy [107][108][109]. In this area, observations are inevitably prone to distortion due to the presence of the atmosphere.…”

Section: Application Of Deep Learning In a Deconvolution Problemmentioning

confidence: 99%

State-of-the-Art Approaches for Image Deconvolution Problems, including Modern Deep Learning Architectures

Makarkin

Bratashov

2021

Micromachines

View full text Add to dashboard Cite

In modern digital microscopy, deconvolution methods are widely used to eliminate a number of image defects and increase resolution. In this review, we have divided these methods into classical, deep learning-based, and optimization-based methods. The review describes the major architectures of neural networks, such as convolutional and generative adversarial networks, autoencoders, various forms of recurrent networks, and the attention mechanism used for the deconvolution problem. Special attention is paid to deep learning as the most powerful and flexible modern approach. The review describes the major architectures of neural networks used for the deconvolution problem. We describe the difficulties in their application, such as the discrepancy between the standard loss functions and the visual content and the heterogeneity of the images. Next, we examine how to deal with this by introducing new loss functions, multiscale learning, and prior knowledge of visual content. In conclusion, a review of promising directions and further development of deconvolution methods in microscopy is given.

show abstract

Deep learning for a space-variant deconvolution in galaxy surveys

Cited by 34 publications

References 59 publications

Superresolution Reconstruction of Severely Undersampled Point-spread Functions Using Point-source Stacking and Deconvolution

Superresolution Reconstruction of Severely Undersampled Point-spread Functions Using Point-source Stacking and Deconvolution

SLITRONOMY: Towards a fully wavelet-based strong lensing inversion technique

State-of-the-Art Approaches for Image Deconvolution Problems, including Modern Deep Learning Architectures

Contact Info

Product

Resources

About