Back to search

University of Southampton

Machine learning for high energy astronomy surveys

Abstract

dc:description.abstract

This thesis presents new machine learning techniques for producing high energy astronomy survey catalogues. A novel source detector is developed for application to images from the INTEGRAL satellite. This source detector utilises convolutional neural networks (CNNs) to confidently identify genuine astrophysical sources whilst rejecting instrumental artefacts. This CNN-based source detector is substantially faster than previous methods, enabling the search for sources on shorter timescales than older techniques used in the production of previous INTEGRAL catalogues. The new capabilities afforded by the CNN source detector resulted in a 5% increase in sources found from the same dataset used to produce the previous INTEGRAL catalogue. A Bayesian source combination technique is also presented that rapidly and reliably combines excess detections into a list of distinct sources. This method is superior to previous approaches because it requires no human intervention, and thus is less prone to human bias. It also is insensitive to the order in which excesses are presented to the algorithm, thereby providing consistent source catalogues regardless of how new detections are included. Finally, a burst detection tool built with long short-term memory (LSTM) networks is presented. This burst detector reliably detects outbursts in simulated data sets (where the ground truth is known) with the same accuracy as previous tools but operating at substantially faster speeds. The burst detector demonstrates potential for applying reliable burst detection to massive data sets like those expected to be produced by the next generation of high energy surveys. Overall, this thesis presents a powerful set of tools that could transform the way high energy astronomy surveys operate. Whilst this thesis demonstrates the advantages of using these tools for catalogue production, they have potential applications in real-time survey operations such as follow up triggers after real-time outburst detection. Tools like those presented here will be vital for high energy astrophysics in the era of big data.

Degree

thesis:*
Name dc:type.qualificationname
Ph.D.
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Southampton
Year dc:date.issued
2023

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Childress, Victoria Adele
Advisors dc:contributor.advisor
  • Gunn, Steve
  • Bird, Tony
ORCID person:identifier.orcid

Rights

Language dc:language
English, eng

Chain of custody

source
Harvested from
University of Southampton
Base URL
eprints.soton.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Childress, Victoria Adele. Machine learning for high energy astronomy surveys. doctoral thesis, University of Southampton, 2023.