Physically constrained causal noise models for high-contrast imaging of exoplanets

TL;DR

This paper introduces a novel domain-knowledge-based causal noise modeling approach for high-contrast exoplanet imaging, improving detection accuracy by integrating scientific understanding with machine learning.

Contribution

It presents a modified half-sibling regression framework that combines domain knowledge with machine learning for enhanced post-processing in exoplanet imaging.

Findings

Outperforms existing leading algorithms in visual quality and SNR on real datasets

Demonstrates the effectiveness of domain-knowledge integration in noise modeling

Potential to enable new exoplanet discoveries in archival data

Abstract

The detection of exoplanets in high-contrast imaging (HCI) data hinges on post-processing methods to remove spurious light from the host star. So far, existing methods for this task hardly utilize any of the available domain knowledge about the problem explicitly. We propose a new approach to HCI post-processing based on a modified half-sibling regression scheme, and show how we use this framework to combine machine learning with existing scientific domain knowledge. On three real data sets, we demonstrate that the resulting system performs clearly better (both visually and in terms of the SNR) than one of the currently leading algorithms. If further studies can confirm these results, our method could have the potential to allow significant discoveries of exoplanets both in new and archival data.