2020A&A...635A..90B

Context. The improvement of large size detectors permitted the development of integral field spectrographs (IFSs) in astronomy. Spectral information for each spatial element of a two-dimensional field of view is obtained thanks to integral field units that spread the spectra on the 2D grid of the sensor.
Aims. Here we aim to solve the inherent issues raised by standard data-reduction algorithms based on direct mapping of the 2D+λ data cube: the spectral cross-talk due to the overlap of neighbouring spectra, the spatial correlations of the noise due to the re-interpolation of the cube on a Cartesian grid, and the artefacts due to the influence of defective pixels.
Methods. The proposed method, Projection, Interpolation, and Convolution (PIC), is based on an "inverse-problems" approach. By accounting for the overlap of neighbouring spectra as well as the spatial extension in a spectrum of a given wavelength, the model inversion reduces the spectral cross-talk while deconvolving the spectral dispersion. Considered as missing data, defective pixels undetected during the calibration are discarded on-the-fly via a robust penalisation of the data fidelity term.
Results. The calibration of the proposed model is presented for the Spectro-Polarimetric High-contrast Exoplanet REsearch instrument (SPHERE). This calibration was applied to extended objects as well as coronagraphic acquisitions dedicated to exoplanet detection or disc imaging. Artefacts due to badly corrected defective pixels or artificial background structures observed in the cube reduced by the SPHERE data reduction pipeline are suppressed while the reconstructed spectra are sharper. This reduces the false detections by the standard exoplanet detection algorithms.
Conclusions. These results show the pertinence of the inverse-problems approach to reduce the raw data produced by IFSs and to compensate for some of their imperfections. Our modelling forms an initial building block necessary to develop methods that can reconstruct and/or detect sources directly from the raw data.