Leuzzi, Laura
(2021)
Characterization of convolutional neural networks for the identification of galaxy-galaxy strong lensing events.
[Laurea magistrale], Università di Bologna, Corso di Studio in
Astrofisica e cosmologia [LM-DM270]
Documenti full-text disponibili:
![[img]](http://amslaurea.unibo.it/style/images/fileicons/application_pdf.png) |
Documento PDF (Thesis)
Disponibile con Licenza: Salvo eventuali più ampie autorizzazioni dell'autore, la tesi può essere liberamente consultata e può essere effettuato il salvataggio e la stampa di una copia per fini strettamente personali di studio, di ricerca e di insegnamento, con espresso divieto di qualunque utilizzo direttamente o indirettamente commerciale. Ogni altro diritto sul materiale è riservato
Download (16MB)
|
Abstract
Studying Galaxy-Galaxy Strong Lensing events allows to tackle several problems, that include the reconstruction of the mass distribution of the lens galaxies and the estimation of the Hubble constant. Thousands of these systems are expected to be detected in upcoming imaging surveys, such as the one that will be carried out by the Euclid space telescope, but they will have to be identified among the billions of sources that will be observed. In this context, the development of automated and reliable techniques for the examination of large volumes of data is of crucial importance. Convolutional Neural Networks are a Deep Learning technique that has proven particularly effective in the past years as a poweful tool for the analysis of large datasets, because of their speed of execution and capacity of generalization. In this thesis work, in particular, we evaluate the ability of this kind of Neural Networks to identify these events on the basis of their morphological characteristics, comparing the performance of three different architectures. For this purpose, we have used two datasets, composed of images simulated to mimic the data quality expected by the observations of the Euclid space telescope: the lenses in the first dataset are characterized by a diverse and complex morphology, while the lenses in the second one are mainly recognizable because of large arcs and rings. Specifically, we have evaluated the performance of the networks on different selections of images, gradually including larger fractions of borderline objects, as well as their ability to identify the most evident lenses. Moreover, we have investigated the possible impact of some of the main characteristics of the lenses and sources on our results. Our analysis has confirmed the potential of the application of this method for the identification of clear lenses, while it has highlighted the need of specific training for the detection of fainter lensing features.
Abstract
Studying Galaxy-Galaxy Strong Lensing events allows to tackle several problems, that include the reconstruction of the mass distribution of the lens galaxies and the estimation of the Hubble constant. Thousands of these systems are expected to be detected in upcoming imaging surveys, such as the one that will be carried out by the Euclid space telescope, but they will have to be identified among the billions of sources that will be observed. In this context, the development of automated and reliable techniques for the examination of large volumes of data is of crucial importance. Convolutional Neural Networks are a Deep Learning technique that has proven particularly effective in the past years as a poweful tool for the analysis of large datasets, because of their speed of execution and capacity of generalization. In this thesis work, in particular, we evaluate the ability of this kind of Neural Networks to identify these events on the basis of their morphological characteristics, comparing the performance of three different architectures. For this purpose, we have used two datasets, composed of images simulated to mimic the data quality expected by the observations of the Euclid space telescope: the lenses in the first dataset are characterized by a diverse and complex morphology, while the lenses in the second one are mainly recognizable because of large arcs and rings. Specifically, we have evaluated the performance of the networks on different selections of images, gradually including larger fractions of borderline objects, as well as their ability to identify the most evident lenses. Moreover, we have investigated the possible impact of some of the main characteristics of the lenses and sources on our results. Our analysis has confirmed the potential of the application of this method for the identification of clear lenses, while it has highlighted the need of specific training for the detection of fainter lensing features.
Tipologia del documento
Tesi di laurea
(Laurea magistrale)
Autore della tesi
Leuzzi, Laura
Relatore della tesi
Correlatore della tesi
Scuola
Corso di studio
Ordinamento Cds
DM270
Parole chiave
astrophysics,lensing,machine,learning,cnn
Data di discussione della Tesi
12 Marzo 2021
URI
Altri metadati
Tipologia del documento
Tesi di laurea
(NON SPECIFICATO)
Autore della tesi
Leuzzi, Laura
Relatore della tesi
Correlatore della tesi
Scuola
Corso di studio
Ordinamento Cds
DM270
Parole chiave
astrophysics,lensing,machine,learning,cnn
Data di discussione della Tesi
12 Marzo 2021
URI
Statistica sui download
Gestione del documento: