I Can't Believe It's Not JWST MIRI

JWST MIRI is already revolutionising mid-infrared astronomy. MIRI observes at wavelengths of 4.8-29 microns, and is designed to distinguish ultra-high-redshift galaxies from more nearby systems, to track the growth of heavily dust-enshrouded supermassive black holes at cosmic noon, and to trace the birth of stars. However, the very small projected areas on the sky of the MIRI detectors means that only very narrow, deep fields can be imaged. Much wider infrared survey data exists, but at a coarser angular resolution.
In the last half a decade Generative Adversarial Networks, denoising auto-encoders and other technologies have been used to attempt deconvolutions on optical data. We have developed an auto-encoder with a novel loss function to overcome this problem in the submillimeter wavelength range (Lauritsen+21 MNRAS, 507, 1546). This approach is successfully demonstrated on European Space Agency Herschel observatory SPIRE 500 micron COSMOS data, with the superresolving target being the James Clerk Maxwell Telescope SCUBA-2 450 micron observations of the same field. We reproduce the SCUBA-2 images with high fidelity using this auto-encoder.
This technique has been very successful in blank-field extragalactic surveys, in which the distant galaxies appear as unresolved point sources. However, we have found that the deconvolution struggles when there is extended structure in the images, due to interstellar cirrus from our Galaxy. This is not unexpected, because the deconvolution training sets did not incorporate this extended emission.