Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Multimodal morphological classification of radio galaxies
Thesis (MSc)--Stellenbosch University, 2025.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Stellenbosch : Stellenbosch University
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867614015819087872 |
|---|---|
| access_status_str | Open Access |
| author | Farge, Dylan Richard |
| author2 | Grobler, Trienko |
| author_browse | Farge, Dylan Richard Grobler, Trienko |
| author_facet | Grobler, Trienko Farge, Dylan Richard |
| author_sort | Farge, Dylan Richard |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MSc)--Stellenbosch University, 2025. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/134615 |
| institution | Stellenbosch University (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:45:19.124Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2025 |
| publishDateRange | 2025 |
| publishDateSort | 2025 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/134615 Multimodal morphological classification of radio galaxies Farge, Dylan Richard Grobler, Trienko Stellenbosch University. Faculty of Science. Dept. of Mathematical Sciences. Computer Science Division. Radio galaxies -- Classification Multimodal user interfaces (Computer systems) Deep learning (Machine learning) Thesis (MSc)--Stellenbosch University, 2025. Farge, D. R. 2025.Multimodal Morphological Classification of Radio Galaxies. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/b33c720d-a579-4923-a1b8-6b0b3905467b ENGLISH ABSTRACT: The morphological classification of radio galaxies is an important task, ulti-mately deepening our understanding of galactic evolution and the interactions of galaxies with their surrounding environments. With advances in observa-tional instruments, the number of observed and/or catalogued radio galaxies is rapidly increasing, creating an acute need for an efficient and scalable classi-fication method. At the same time, increased inter and intra institute collabo-ration, in addition to the development of multiple new facilities – each with its own observational specifications – provide unique unexplored opportunities to leverage various modalities of data for training Deep Neural Networks (DNNs) to perform the morphological classification task. This raises the question of how advantageous multimodal datasets will be, and what is the most effective strategy of employing them to execute said task. In this thesis, images from three radio surveys with distinct observational attributes were combined to construct the Radio Analysis of Duplicates Cata-logue (RADCAT) dataset. Raw unimaged interferometric measurements were also re-engineered from these same images to create a separate accompanying dataset. These two datasets were then employed to train various deep learning models to discern between various radio galaxies. Moreover, two distinct neural network architectures were evaluated using the raw interferometric data, namely a Convolutional Neural Network (CNN) and a Recurrent Neural Network (RNN). The use of multiple data modalities was found to significantly improve model stability and robustness, regardless of whether image processing was utilised. Preliminary comparisons between the efficacy of DNNs, trained on raw measurements and traditional image-based inputs, suggest that raw data holds promise as an alternative input represen-tation. Models trained with raw data, however, were often less performant than those trained on conventional images, but performant enough to war-rant further investigation. As for architectural performance, the preliminary study conducted in this thesis found that the well-adopted CNN architecture readily outperformed the simple RNN architecture. This finding needs to be confirmed with further follow-up studies, considering a more comprehensive form of model capacity exploration. Furthermore, this thesis also introduces a novel approach to catalogue merging. The proposed approach reduces the need for manual inspection by using a graph based approach to automatically resolve complicated duplicate relationships among proximate sources. Such an approach is particularly valu-able as increasingly sensitive instruments generate ever more contiguous source catalogues. AFRIKAANSE OPSOMMING: Die morfologiese klassifikasie van radiosterrestelsels is ’n belangrike taak. Die taak dra daartoe by dat ons kennis aangaande sterrestelsels en die interaksies van sterrestelsels met hul omliggende omgewings verdiep. Met die vooruitgang in waarnemingsinstrumente neem die aantal waargenome en/of gekatalogiseerde radiosterrestelsels vinnig toe. Hierdie groot hoeveelheid radiosterrestelsesl kan slegs doeltreffend geklassifiseer word met behulp van ’n automatiese en skaleerbare metode. Tegelykertyd bied verhoogde inter- en intra-institusionele samewerking, sowel as die ontwikkeling van verskeie nuwe fasiliteite – elk met sy eie waarnemingspesifikasies – unieke, nog onontginde moontlikhede. Moontlikhede soos die benutting van verskeie datamodaliteite waarmee Diep Neurale Netwerke (DNNs) afgerig kan word om die morfologiese klassifikasietaak uit te voer. Dit bring die volgende vrae na vore, hoe voordelig sal multimodale datastelle wees, en wat is die mees doeltreffendste strategie wat ingespan kan word om die alreeds genoemde taak uit te voer. In hierdie proefskrif is beelde van drie radiohemelruimopnames met onderskeie waarnemingskenmerke gekombineer om die Radio-analise van Duplikaatkatalogus (RADCAT) te konstrueer. Rou, onafgebeelde interferometriese metings is ook herskep uit dieselfde beelde om ’n aparte, aanvullende datastel te maak. Hierdie twee datastelle is toe gebruik om verskeie diep-leer modelle af te rig om tussen verskillende radiosterrestelsels te onderskei. Verder is twee onderskeie neurale netwerk-argitekture geëvalueer deur gebruik te maak van die rou interferometriese data, naamlik ’n Konvolusionêre Neurale Netwerk (CNN) en ’n Herhalende Neurale Netwerk (RNN). Die gebruik van veelvuldige datamodaliteite is gevind om modelstabiliteit en robuustheid aansienlik te verbeter, ongeag of vooraf beeldverwerking toegepas is of nie. Voorlopige vergelykings tussen die doeltreffendheid van DNNs wat op rou meetings afgerig is en dié wat op tradisionele beeldgebaseerde insette opgelei is, dui daarop dat rou data potensiaal inhou as ’n alternatiewe insetvoorstelling. Modelle wat met rou data opgelei is, was egter dikwels minder doeltreffend as dié wat op konvensionele beelde opgelei is, maar steeds doeltreffend genoeg om verdere ondersoek te regverdig. Wat argitektoniese werkverrigting betref, het die voorlopige studie wat in hierdie proefskrif uitgevoer is, bevind dat die goédaanvaarde CNN-argitektuur die eenvoudige RNN-argitektuur duidelik oortref het. Hierdie bevinding moet bevestig word deur verdere opvolgstudies wat ’n meer omvattende vorm van modelkapasiteitsondersoek in ag neem. Hierdie proefskrif bied verder ’n nuwe metode vir die samevoeging van katalogusse aan. Die voorgestelde benadering verminder die behoefte aan handmatige inspeksie deur ’n grafiek-gebaseerde metode te gebruik om ingewikkelde duplikaatverhoudings tussen nabygelëe bronne outomaties op te los. Só ’n benadering is veral waardevol aangesien toenemend sensitiewe instrumente al hoe meer ruimtelik dig bronkatalogusse genereer. Masters 2025-12-18T13:21:45Z 2025-12-18T13:21:45Z 2025-12 Thesis https://scholar.sun.ac.za/handle/10019.1/134615 en Stellenbosch University xvii, 125 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Radio galaxies -- Classification Multimodal user interfaces (Computer systems) Deep learning (Machine learning) Farge, Dylan Richard Multimodal morphological classification of radio galaxies |
| title | Multimodal morphological classification of radio galaxies |
| title_full | Multimodal morphological classification of radio galaxies |
| title_fullStr | Multimodal morphological classification of radio galaxies |
| title_full_unstemmed | Multimodal morphological classification of radio galaxies |
| title_short | Multimodal morphological classification of radio galaxies |
| title_sort | multimodal morphological classification of radio galaxies |
| topic | Radio galaxies -- Classification Multimodal user interfaces (Computer systems) Deep learning (Machine learning) |
| url | https://scholar.sun.ac.za/handle/10019.1/134615 |
| work_keys_str_mv | AT fargedylanrichard multimodalmorphologicalclassificationofradiogalaxies |