Abstract :
[en] Quasars are one of the most peculiar types of objects in astronomy. The supermassive black hole these harbour effectively makes the surrounding matter radiates an enormous amount of energy before getting in the vicinity of the black hole horizon out of which it will never escape. This ironically leads to the most luminous phenomenon in the Universe while being non-transient. It is hence quite natural to rely on these cosmic headlights, visible up to ages when the Universe was still very young, so as to achieve some of the currently most important cosmological applications, notably regarding the determination of the cosmological parameters $H_0$, $\Omega_\Lambda$ and $\Omega_m$. The Gaia mission, on its side, is one of a kind given the one billion of celestial objects it is intended to observe, among which more than half a million quasars are expected. Furthermore, owing to its exceptional astrometric precision, Gaia stands out to be extremely well suited for the detection of gravitational lens (GL) systems. In the latter, light rays coming from a distant background quasar are deflected by the presence of a massive galaxy being in the line-of-sight that leads to the production of multiple images of this background quasar upon a favourable alignment between the quasar, the galaxy and the observer. Supplemental constraints on the aforementioned cosmological parameters being then gained based on these GLs. Gaia hence provides an unprecedented opportunity to detect and characterize quasars as well as to identify GLs which ultimately bring a better understanding of the Universe we live in. This thesis is accordingly concerned with the development of software solutions dedicated to the determination of the astrophysical parameters (APs) of the quasars that Gaia will observe, on one hand, and to the recognition of the GLs among the billion of sources it will uncover, on the other hand.
Although Gaia provides state-of-the-art astrometric and photometric observations, its capability in characterizing these celestial objects remains however restricted by the relatively low spectral resolution of the blue and red spectrophotometers upon which it is based as well as by the limited signal-to-noise ratio that is associated with faint objects, including quasars. In addition, the overwhelming amount of data that Gaia has to process translates into a stringent need for algorithms having both low numerical complexities as well as low memory usages. These restrictions and shortcomings along with the requirement for reliable APs were at the heart of this research that led to the development of two specifically designed methods that are the weighted principal components analysis and the weighted phase correlation method. The former of these methods allowed us to extract the most significant patterns out of quasars with a view of using these in the production of a spectral library of quasars as observed by Gaia. These were subsequently used in a fast and automated procedure designed to guess the redshift of the quasars within the Gaia mission through the latter mentioned method. Other APs that are the slope of the quasar continua, the total equivalent width of their emission lines and whether these encompass broad absorption lines or not, being then concurrently derived based on the results of these methods.
Finally, the identification of GL candidates relies on the recognition of the structures and symmetries that are observed within lensed images through supervised learning methods. The specific method we choose to use, based on extremely randomized trees, was shown to yield a low contamination rate on simulated configurations composed of three images as well as a very high probability of detection in cases of four image configurations. Real observations out of the first Gaia data release were processed and resulted in the identification of candidates having three potentially lensed images which are currently waiting for confirmation using ground-based facilities.
