Thesis Detección de galaxias oscuras utilizando métodos de deep learning
Loading...
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Informática
Departament
Campus
Campus Santiago San Joaquín
Abstract
El efecto de lente gravitacional permite observar y estudiar cuerpos no visibles, como los halos de materia alrededor de la parte luminosa de la galaxia. Sin embargo, no existe una gran cantidad de casos aceptados en los cuales este efecto ocurre cuando la lente no es visible. Por esto en este trabajo se plantea generar imágenes de lensing donde la lente es un agente oscuro utilizando deeplenstronomy y utilizar estas imágenes para entrenar un modelo de deep learning.
Los resultados obtenidos muestran una serie de imágenes realistas y representativas, algunas de las cuales fueron evaluadas por un experto, quien también ha identificado un cierto porcentaje de casos que presentan características menos consistentes con las observaciones. Además, se observó que el modelo entrenado fue capaz de diferenciar los efectos de lente oscuro con los efectos de lente brillante, sin embargo no tuvo tanto problema de diferenciar los casos positivos de lensing y los caso negativos cuando se quitaban los casos brillantes.
The gravitational lensing effect allows us to observe and study non-visible objects, as the halos of matter around the luminous part of the galaxy. However, there are not many accepted cases where this effect occurs when the lens is not visible. Therefore, this study proposes generating lensing images where the lens is an dark agent using deeplenstronomy and using these images to train a deep learning model. The obtained results show a series of realistic and representative images, some of which evaluated by an expert, who has also identified a certain percentage of cases that exhibit characteristics less consistent with the observations. Furthermore, it was observed that the trained model was able to differentiate between the effects of dark and bright lensing systems, but it had less difficulty distinguishing between positive lensing cases and negative cases when the bright cases were removed.
The gravitational lensing effect allows us to observe and study non-visible objects, as the halos of matter around the luminous part of the galaxy. However, there are not many accepted cases where this effect occurs when the lens is not visible. Therefore, this study proposes generating lensing images where the lens is an dark agent using deeplenstronomy and using these images to train a deep learning model. The obtained results show a series of realistic and representative images, some of which evaluated by an expert, who has also identified a certain percentage of cases that exhibit characteristics less consistent with the observations. Furthermore, it was observed that the trained model was able to differentiate between the effects of dark and bright lensing systems, but it had less difficulty distinguishing between positive lensing cases and negative cases when the bright cases were removed.
Description
Keywords
Efecto de lente, Sérsic, Lente oscuro
