Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
FPGA accelerated radio galaxy classification with a Fourier convolutional neural network
Edwards, B. J. 2025. FPGA Accelerated Radio Galaxy Classification with a Fourier Convolutional Neural Network. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/0f7df7f9-5856-46d0-819c-c7f9bf4f8335
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Stellenbosch : Stellenbosch University
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613752842518528 |
|---|---|
| access_status_str | Open Access |
| author | Edwards, Barend Jacobus |
| author2 | Barnard, Arno |
| author_browse | Barnard, Arno Edwards, Barend Jacobus |
| author_facet | Barnard, Arno Edwards, Barend Jacobus |
| author_sort | Edwards, Barend Jacobus |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Edwards, B. J. 2025. FPGA Accelerated Radio Galaxy Classification
with a Fourier Convolutional Neural Network. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/0f7df7f9-5856-46d0-819c-c7f9bf4f8335 |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/132197 |
| institution | Stellenbosch University (South Africa) |
| last_indexed | 2026-06-10T12:41:07.950Z |
| 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:10019.1/132197 FPGA accelerated radio galaxy classification with a Fourier convolutional neural network Edwards, Barend Jacobus Barnard, Arno Grobler, T. L. Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. Radio galaxies -- Classification Neural networks (Computer science) Field programmable gate arrays Machine learning -- Mathematical models UCTD Edwards, B. J. 2025. FPGA Accelerated Radio Galaxy Classification with a Fourier Convolutional Neural Network. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/0f7df7f9-5856-46d0-819c-c7f9bf4f8335 Thesis (MEng)--Stellenbosch University, 2025. ENGLISH ABSTRACT: Morphological classification of observed radio sources has become fundamental in studies related to galaxy formation and evolution. It is therefore important to modern astronomy as a whole. Recently, Convolutional Neural Networks (CNNs) have enabled researchers to implement network architectures that are able to classify radio galaxies with high accuracies. However, the spatial convolutions in these networks are computationally expensive. This project aims to replace the spatial convolutions in a CNN with elementwise multiplications in the frequency domain to reduce computational complexity. We then investigate speeding up inference and reducing hardware power consumption through the use of an Alveo U280 Field Programmable Gate Array (FPGA) accelerator card. We developed a frequency domain CNN capable of morphological galaxy classification. Our proposed model requires fewer computations for classification than tested radio galaxy CNN architectures with a similar number of trainable parameters. However, we found that this computational reduction comes at the cost of a lower accuracy. We were able to speed up inference and improve the power efficiency of our proposed model with the FPGA accelerator card compared to a Central Processing Unit (CPU)-only implementation of the model. AFRIKAANSE OPSOMMING: Die morfologiese klassifisering v a n r a diobronne het fundamenteel geword in studies rakende die vorming en evolusie van sterrestelsels. Dit is dus belangrik vir moderne sterrekunde as geheel. Die onlangse verwikkelinge in konvolusionele neurale netwerke (KNNe) stel navorsers in staat om radio-sterrestelsels met hoë akkuraatheid te kan klassifiseer. Die konvolusies in hierdie netwerke is egter berekening-intensief. Hierdie projek beoog om die spasiële konvolusies in sogenaamde netwerke te vervang met elementgewys vermenigvuldigings in die frekwensiedomein sodat die totale hoeveelheid bewerkings wat nodig is vir klassifisering v erminder k an w o rd. S aam hi ermee b eoog di e projek o m di e sp oed van klassifisering te verhoog en die energieverbruik daarvan te verminder deur die gebruik van ‘n Alveo U280 FPGA-versnellingskaart. Ons het ‘n frekwensiedomein-KNN ontwikkel wat in staat is tot morfologiese klassifikasie van radio-sterrestelsels. Die voorgestelde model vereis minder berekeninge vir klassifisering teenoor getoetsde radio-sterrestelsel-KNN-argitekture met soortgelyke opleibare parameters. Ons het egter gevind dat die vermindering in berekeninge ten koste van ‘n laer akkuraatheid kom. Met die gebruik van die FPGA-versnellingskaart kon ons die spoed van klassifisering e n d ie k ragdoeltreffendheid va n di e vo orgestelde model ve rbeter in vergelyking met ‘n sentrale verwerkingseenheid (SVE)-implementering van die model. Masters 2025-05-29T10:31:20Z 2025-05-29T10:31:20Z 2025-03 Thesis https://scholar.sun.ac.za/handle/10019.1/132197 Stellenbosch University xvii, 109 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Radio galaxies -- Classification Neural networks (Computer science) Field programmable gate arrays Machine learning -- Mathematical models UCTD Edwards, Barend Jacobus FPGA accelerated radio galaxy classification with a Fourier convolutional neural network |
| title | FPGA accelerated radio galaxy classification with a Fourier convolutional neural network |
| title_full | FPGA accelerated radio galaxy classification with a Fourier convolutional neural network |
| title_fullStr | FPGA accelerated radio galaxy classification with a Fourier convolutional neural network |
| title_full_unstemmed | FPGA accelerated radio galaxy classification with a Fourier convolutional neural network |
| title_short | FPGA accelerated radio galaxy classification with a Fourier convolutional neural network |
| title_sort | fpga accelerated radio galaxy classification with a fourier convolutional neural network |
| topic | Radio galaxies -- Classification Neural networks (Computer science) Field programmable gate arrays Machine learning -- Mathematical models UCTD |
| url | https://scholar.sun.ac.za/handle/10019.1/132197 |
| work_keys_str_mv | AT edwardsbarendjacobus fpgaacceleratedradiogalaxyclassificationwithafourierconvolutionalneuralnetwork |