Convolutional Neural Networks for Tomographic Imaging Spectroscopy of the Solar Atmosphere

Abstract:

The EUV Snapshot Imaging Spectrograph (ESIS) is an tomographic imaging spectrograph designed to measure spectral line profiles over a 2D field-of-view with much faster cadence than a rastering slit spectrograph. ESIS uses four independent slitless spectrographs, each with a different dispersion direction but all fed from the same primary mirror. To recover spectral line profiles from this arrangement, the images from each slitless spectrograph are interpreted using computed tomography algorithms. With only four independent spectrographs, this is a classic limited-angle tomography problem. We trained a convolutional neural network to solve this tomography problem using observations from the Coronal Diagnostic Spectrometer (CDS) as a training dataset. We will present the performance of this network along with its application to the observations gathered during the 2019 ESIS sounding rocket flight.