Penchev, Petar (2025) A Deep Learning approach for removing AGN contributions from galaxy observations. Bachelor's Thesis, Astronomy.
|
Text
BachelorsThesisDeepRemovalofAGNDRAGN.pdf Download (13MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
Toestemming.pdf Restricted to Registered users only Download (217kB) |
Abstract
Supermassive black holes (SMBHs) and their associated active galactic nuclei (AGN) play a pivotal role in galaxy evolution, yet their intense luminosity often obscures key structural details of their host galaxies. Traditional AGN subtraction techniques rely on parametric modeling tools such as GALFIT, which, while effective, are computationally expensive, inflexible, and require manual fine-tuning. In this thesis, we present a novel, data-driven alternative: DRAGN (Deep Removal of AGN), a deep learning framework for removing AGN contributions from galaxy images using convolutional neural networks (CNNs). We investigate three distinct architectures—U-Net, Attention U-Net, and conditional Generative Adversarial Network (cGAN)—evaluating their ability to perform image-to-image (i2i) translation from AGN-contaminated to AGN-free galaxy images. A training dataset of over 500,000 simulated galaxy-AGN pairs, constructed to mimic the JWST/NIRCam F150W filter, is used to develop the models. Our results show that the U-Net outperforms both its attention-enhanced variant and the cGAN in terms of image fidelity, spatial accuracy, and flux recovery, making it the most well-balanced mode. This study establishes DRAGN as a scalable and effective alternative for AGN subtraction, with the potential to revolutionize morphological studies of AGN host galaxies in current and upcoming deep sky surveys.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Wang, L. and Koopmans, D.M. |
| Degree programme: | Astronomy |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 14 Jul 2025 10:02 |
| Last Modified: | 14 Jul 2025 10:02 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/36121 |
Actions (login required)
 |
View Item |