• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Master's Dissertation
DOI
https://doi.org/10.11606/D.14.2023.tde-22082023-115704
Document
Author
Full name
Leonardo Vieira Costa
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Paulo, 2023
Supervisor
Committee
Cypriano, Eduardo Serra (President)
Dupke, Renato de Alencar
Sodre Junior, Laerte
Vitorelli, André Zamorano
Title in Portuguese
Métodos de medição de forma de galáxias livre da PSF para o levantamento J-PAS.
Keywords in Portuguese
Galáxias
Inteligência artificial
J-PAS
Lentes gravitacionais
Medida de forma
Redes neurais convolucionais.
Abstract in Portuguese
A partir dos dados do levantamento mini J-PAS, que são os primeiros dados públicos do J-PAS, apresentamos uma otimização da modelagem da PSF usando os {\it softwares} {\sc SExtractor} e {\sc PSFEx}, mais uma série de testes de qualidade do modelo da PSF, como as estatísticas de Rowe e outros testes nulos, que atestam a validade do nosso modelo. Usando o módulo HSM do {\sc GalSim}, escolhemos o método KSB com Re-Gaussianização para calcular os valores de cisalhamento, dos quais serão usados na calibração da relação massa-observável para contagens de aglomerados de galáxias. Além desse método, usamos técnicas de rede neural convolucional (CNN) para obter valores de cisalhamento corrigidos pela PSF a partir de uma dada imagem da galáxia. Comparamos nossos resultados de cisalhamento com um levantamento mais profundo, o CFHTLenS. Usando o KSB com regaussianização e nossa CNN baseada em uma \textit, obtemos um coeficiente de correlação de Pearson de $\sim0.86$ e $\sim0.88-0.90$, respectivamente. Entretanto, não houve melhoras em incorporar a PSF na CNN. Por fim, discutimos os desafios de introduzir correções da PSF no contexto das CNNs e quais técnicas de aprendizado de máquina devem ser usadas em seu lugar.
Title in English
PSF free galaxy shape measurement methods for the J-PAS survey
Keywords in English
Artificial intelligence
Convolutional neural networks
Galaxies
Gravitational lensing
J-PAS
Shape measurement
Abstract in English
From the mini J-PAS survey data, which is the first public data of the J-PAS, we present an optimization of the PSF modeling using the {\it softwares} {\sc SExtractor} and {\sc PSFEx}, plus a series of PSF model quality tests, such as the Rowe statistics and other null tests, which attest to the validity of our model. Using the {\sc GalSim} HSM module, we choose the KSB Re-Gaussianization method to compute the shear values, which will be used in the calibration of the mass-observable relation for Galaxy cluster counts. In addition to this method, we use convolutional neural network (CNN) techniques to obtain PSF-corrected shear values from given galaxy images. We compare our shear results with a deeper survey, the CFHTLens. Using the KSB Re-Gaussianization and our CNN, we obtain a Pearson correlation coefficient of $\sim0.86$ and $\sim0.88-0.90$, respectively. However, there was no improvement in incorporating the PSF into CNN. Finally, we discuss the challenges of introducing PSF corrections in the context of CNNs and which machine learning techniques should be used instead.
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
Corrigida.pdf (8.30 Mbytes)
Publishing Date
2023-09-05
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.