Back to search

The Open University

Super-resolving <i>Herschel</i> SPIRE images using Convolutional Neural Networks

Abstract

dc:description.abstract

Wide-field sub-millimetre surveys have allowed many advances in galaxy evolution, but their lower angular resolution, compared to optical surveys, limits the science exploitation. Consequently, various deconvolution methods have been developed. In the last decade, Generative Adversarial Networks have been increasingly developed and used to attempt deconvolutions on optical data. This thesis presents an U-Net with a novel loss function to provide super-resolution on sub-millimetre observations. This approach is successfully demonstrated on <i>Herschel</i> SPIRE 500μm COSMOS data, with the super-resolving target being the corresponding JCMT SCUBA-2 450μm images. The JCMT SCUBA-2 image information is reproduced with good accuracy in both the point source positions and fluxes and good completeness and purity. This roughly doubles the de-convolved area in the COSMOS field compared to previous de-convolution work using multi-wavelength priors. The super-resolution results are cross-matched to existing mid-infrared and sub-millimetre surveys by associating them with their nearest neighbour and also using a statistically motivated association method referred to in this thesis as the likelihood ratio method. The associated source fluxes are either based on the cross-matched catalogue fluxes or calculated using forced photometry at the super-resolved position. These results are used to calculate associated photometric redshifts. Multiplicity calculations are made on the COSMOS field using the super-resolution results from this thesis, alternative de-convolution results provided by the <i>Herschel</i> Extragalactic Legacy Project, and the JCMT SCUBA-2 STUDIES survey. These are compared to theoretical predictions from the Empirical Galaxy Generator. Finally, foreground cirrus contamination is found to be the limiting factor in the expansion of the super-resolution algorithm to the fields studied by the JCMT SCUBA-2 RAGERS project. However, a boxcar background subtraction filter is developed that is partially successful in removing cirrus contamination and in improving the super-resolution.

Degree

thesis:*
Name dc:type.qualificationname
phd
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
The Open University
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Lauritsen, Lynge Roald Bunk

Rights

Language dc:language
en

Chain of custody

source
Harvested from
The Open University
Base URL
oro.open.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Lauritsen, Lynge Roald Bunk. Super-resolving <i>Herschel</i> SPIRE images using Convolutional Neural Networks. doctoral thesis, The Open University, 2025.