Reference : Advanced data processing for high-contrast imaging - Pushing exoplanet direct detecti...
Dissertations and theses : Doctoral thesis
Physical, chemical, mathematical & earth Sciences : Space science, astronomy & astrophysics
http://hdl.handle.net/2268/214337
Advanced data processing for high-contrast imaging - Pushing exoplanet direct detection limits with machine learning
English
Gómez González, Carlos mailto [Université de Liège - ULiège > Département d'astrophys., géophysique et océanographie (AGO) > Astroph. extragalactique et observations spatiales (AEOS) >]
13-Sep-2017
Université de Liège, ​Liège, ​​Belgique
Doctor of Science
Surdej, Jean mailto
Van Droogenbroeck, Marc mailto
Absil, Olivier mailto
Wehenkel, Louis mailto
Gillon, Michaël mailto
Marois, Christian mailto
Quanz, Sascha mailto
Mawet, Dimitri mailto
[en] Exoplanets ; Image processing ; High-contrast imaging ; Machine learning ; Deep learning
[en] Since ancient times, mankind has wondered whether other solar systems exist around other stars somewhere in the Universe. It took many centuries to finally prove the existence of extra-solar planetary systems. Nowadays, more than 3500 exoplanets have been discovered, mostly thanks to indirect detection methods. Indeed, the task of directly detecting exoplanets through high-contrast imaging (HCI) is a formidable challenge, and has only been enabled in the last decade thanks to advances in instrumentation and dedicated image processing algorithms. This last component of the exoplanet direct imaging pipeline is what ultimately pushes the detection limits and sensitivity of HCI instruments and survey campaigns. Unfortunately, the HCI community has been slow in adopting the latest developments in data management and machine learning for analyzing the increasing amount of available data. This dissertation is an attempt to fill in this very gap, and develops at the interface of computer science, machine learning, statistics, and astrophysics. This work contributes to the field of data processing for HCI in two main ways. On one hand, I have developed an open source \texttt{Python} library for taking HCI data from the raw state up to the characterization of companions. It implements state-of-the-art approaches and is positioning itself as one of the de facto software solutions for building HCI pipelines. I have also participated to the critical analysis of data from different first and second generation HCI instruments. On the other hand, I have approached the task of exoplanet detection in angular differential imaging sequences from a computer vision and machine learning perspective. This interdisciplinary work has led to novel algorithmic solutions, extending unsupervised learning techniques widely used in HCI and proposing advanced supervised learning approaches based on cutting-edge deep learning models. My novel algorithms have been presented using a robust performance assessment framework to produce large comparative performance studies. These studies show the improved sensitivity vs specificity trade-off of the proposed supervised detection approach. The proposed algorithms bring the possibility of re-processing existing HCI databases to maximize their scientific return and potentially improve the demographics of directly imaged exoplanets.
Space sciences, Technologies and Astrophysics Research - STAR
http://hdl.handle.net/2268/214337

File(s) associated to this reference

Fulltext file(s):

FileCommentaryVersionSizeAccess
Open access
GomezGonzalez_Thesis_submitted.pdfAuthor preprint41.63 MBView/Open

Bookmark and Share SFX Query

All documents in ORBi are protected by a user license.