Application of Neural Networks to Searching for Optical Transients in Astronomical Images Using the Subtraction Method

With the beginning of a new gravitational-wave era in astronomy and astrophysics, the problem of identification of gamma-ray bursts optical counterparts is becoming more relevant than ever before. The application of existing methods for identifying transients, in particular image subtraction, is complicated by several factors; the main ones are the variability of observation conditions and the variability of other sources not related to the desired counterpart, as well as the possible absence of reference images from the observatory at which the search is being conducted. This paper examines the results of using deep image prior neural network to translation images from the Pan-STARRS survey to images from the AS-32 telescope of the Abastumani Astrophysical Observatory for the purpose of their subsequent subtraction for the identification and photometry of optical transients. Using examples of fragments of images of the M82 galaxy, the potential of the deep prior approach is demonstrated for both detection of transient events in M82 and for their flux estimation.