Multiple reference star differential imaging with VLT/SPHERE

Influence of the reference frame selection and library^★

¹ Univ. Grenoble Alpes, CNRS, IPAG, 38000 Grenoble, France
² European Southern Observatory, Alonso de Córdova 3107, Vitacura, Casilla 19001, Santiago de Chile, Chile
³ LESIA, Observatoire de Paris, PSL Research University, CNRS, Sorbonne Universités, UPMC Univ. Paris 06, Univ. Paris Diderot, Sorbonne Paris Cité, France
⁴ Department of Astronomy and Steward Observatory, University of Arizona, 933 N. Cherry Avenue, Tucson, AZ 85721-0065, USA
⁵ Large Binocular Telescope Observatory, University of Arizona, 933 N. Cherry Avenue, Tucson, AZ 85721-0065, USA
⁶ Instituto de Astrofísica de Canarias, Calle Vía Láctea S/N, 38205 La Laguna, Tenerife, Spain
⁷ Instituto Nacional de Astrofésica, Optica y Electrónica, Luis Enrique Erro 1, Sta. Ma. Tonantzintla, Puebla, Mexico

^★★ Corresponding author; cromero@eso.org

Received: 9 March 2023
Accepted: 30 August 2024

Abstract

Context. High-contrast imaging observations mostly rely on angular differential imaging, a successful technique for detecting point-sources, such as planets. However, in the vicinity of the star (typically below 300 mas), this technique suffers from signal self-subtraction when there is not enough field rotation. Building large libraries of reference stars from archival data later used to optimally subtract the stellar halo is a powerful technique known as reference star differential imaging (RSDI) that can overcome this limitation.

Aims. We aim at investigating new methods for creating reference libraries composed of multiple stars when applying reference star differential imaging to VLT/SPHERE data. We used for that purpose a data set from the SPHERE High Angular Resolution Debris Disk Survey (SHARDDS), composed of 55 targets observed in broad-band H with the InfraRed Dual-band Imager and Spectrograph (IRDIS) during 2015-2016, with a total of ~20 000 frames. We consider HD 206893, known to host a close-in bound substellar companion HD 206893 B, as a benchmark science target to demonstrate the improved sensitivity provided by this method.

Methods. We created libraries of reference frames based on different image similarity metrics: the cosine distance between descriptors created by a convolutional neural network, the Pearson correlation coefficient, the Structural Similarity Index, the Strehl ratio, and raw contrast criteria. We used principal component analysis (PCA) to subtract the stellar halo and tested various normalization options.

Results. We obtained the best signal-to-noise ratio (S/N) on HD 206893 B by using the Pearson correlation coefficient (PCC) applied to an annulus between 245 and 612 mas to select reference frames. The ten reference libraries with the highest S/N on the substellar companion HD 206893 B were all based on the PCC method, outperforming other similarity metrics. While the Strehl ratio is the environment variable most correlated to the contrast, it is insufficient to select similar images. We also show that having multiple reference stars in the reference library produces better results than using a single well-chosen reference star.

Conclusions. Using the Pearson correlation computed on a specific area of interest to select reference frames is a promising alternative to improve the detectability of faint point-sources when applying reference star differential imaging. In the future, reducing all the data available in the SPHERE archive using this technique might offer interesting results in the search for previously undetected planets.