Abstract :
[en] High contrast imaging (HCI) is one of the most challenging techniques for exoplanet detection, but also one of the most promising. The main difficulties encountered with HCI arise from the small angular separation between the host star and the potential exoplanets, the flux ratio between them, and the image degradation caused by the Earth's atmosphere. Adaptive optics and coronagraphic techniques are now widely used to improve the quality and the dynamic range of the images with dedicated instruments. However, despite the use of these cutting-edge technologies, the resulting images are still affected by residual aberrations. Under good observing conditions, the performance of HCI instruments is limited by aberrations arising in the optical train of the telescope and instrument, generating quasi-statics speckles in the field of view. Different post-processing techniques along with observing strategies have been proposed in the last decade to deal with these quasi-static speckles, whose shape and intensity are similar to potential companions.This PhD thesis builds upon these recent advances, focusing mainly on the development of a new data processing technique to unveil fainter planetary signals from angular differential imaging (ADI) sequences, and to retrieve their observed properties.
Most post-processing techniques are based on the ADI observing strategy and perform a subtraction of a reference point spread function (PSF), which models the speckle field. Such techniques generally make use of signal-to-noise maps to infer the existence of planetary signals via thresholding. An alternative method to generate the final detection map based on a regime-switching model (RSM) is developed in the first part of this thesis. This approach considers a planetary regime and a speckle regime to describe, via a Markov chain, the evolution of the pixels intensity within cubes of residuals generated by one or multiple PSF-subtraction techniques. The short memory process used in the RSM algorithm allows quasi-static speckles to be treated more effectively. Using multiple PSF-subtraction techniques helps reducing further the residual speckle noise level, better discriminating planetary signals from residual speckles. The RSM map algorithm showed an overall better performance in the receiver operating characteristic space when compared with standard signal-to-noise ratio maps for several state-of-the-art ADI-based post-processing algorithms.
Building on the good results obtained with the RSM algorithm, several improvements of the vanilla RSM map algorithm are then implemented. We started by considering two forward-model versions of the RSM map algorithm based on the LOCI and KLIP PSF-subtraction techniques, allowing to account for the planetary signal self-subtraction observed at short separations. We then addressed the question of optimally selecting the PSF subtraction techniques to optimise the overall performance of the RSM map. A new forward-backward approach is also implemented to take into account both past and future observations to compute the RSM map probabilities, leading to improved precision in terms of astrometry and lowering the background speckle noise. Performance analysis demonstrate the benefits of these improvements.
Following these developments, the RSM map algorithm can use up to seven PSF-subtraction techniques. The selection of the optimal parameters for these PSF-subtraction techniques as well as for the RSM map is therefore not straightforward, time consuming, and can be biased by assumptions made as to the underlying data set. We propose in the fourth chapter of this thesis a novel optimisation procedure that can be applied to each of the PSF-subtraction techniques alone, or to the entire RSM framework. This optimisation procedure, called auto-RSM, consists of three main steps: (i) definition of the optimal set of parameters for the PSF-subtraction techniques, (ii) optimisation of the RSM algorithm, and (iii) selection of the optimal set of PSF-subtraction techniques and ADI sequences used to generate the final RSM probability map. The optimisation procedure is applied to the data sets of the exoplanet imaging data challenge (EIDC). The results demonstrate the interest of the proposed optimisation procedure, with better performance metrics compared to the earlier version of RSM, as well as to other HCI data-processing techniques.
The auto-RSM framework is finally applied to the SHARDDS survey to bring an additional piece to the exoplanet puzzle, by contributing to the characterisation of planetary population via the estimation of occurrence rate maps. This survey gathers 55 main-sequence stars within 100\,pc, known to host a high-infrared-excess debris disk, allowing us to potentially better understand the complex interactions between substellar companions and disks. A clustering approach is used to divide the set of targets into multiple subsets, in order to reduce the computation time by estimating a single optimal parametrisation for each considered subset. A new planetary characterisation algorithm, based on the RSM framework, is developed and tested successfully. We uncover the companion around HD206893, but do not detect any new companion around other stars. Planet detection and planet occurrence frequencies are nevertheless derived from the generated contrast curves and show a high sensitivity between 10 and 100 au for substellar companions with masses over 10 Jupiter masses.
Throughout the different chapters of this thesis, we have built a complex but highly efficient post-processing framework for ADI sequences, adding in each chapter many new features and simplifying its use. All these developments have been compiled into a python package, called PyRSM, which offers a parameter-free detection map computation algorithm with a very low level of residual speckles. This package has largely increased in maturity thanks to the SHARDDS survey and has become a robust HCI post-processing pipeline, achieving good performance in terms of contrasts. PyRSM will hopefully be used for many more surveys and provide unprecedented detection limits, allowing the detection of many exoplanets with the next generation of telescopes and instruments.
