Analysis and Benchmarking of Astronomical Transient categorization using Deep Learning Methods

We provide an approach for the detection of optical transients that is based on Deep Learning. We explain the design of two networks that can compare photos of the same portion of the sky captured by various telescopes. One image is from the time when a possible transient could exist, and the other is a reference image from a time before that. We use information from the Dr. Cristina V. Torres Memorial Astronomical Observatory and old reference pictures from the Sloan Digital Sky Survey. After training a convolutional neural network and a dense layer neural network using simulated source samples, we evaluated the performance of the trained networks on samples that were produced using actual picture data. The traditional approach of identifying transients, which generally involves the source extraction of a difference picture followed by human assessment of the candidates found, has been replaced by methods of autonomous detection. If humans were not involved in the inspection process, then a totally automated system might be used instead. This would make it possible to carry out fast and automatic follow-up on intriguing targets of opportunity. The model pipeline that we provide here is not yet capable of replacing human inspection; nonetheless, it may give helpful indications that may be used to identify possible candidates. This method will be improved upon and put to the test with the help of telescopes that are a part of the Transient Optical Robotic Observatory of the South Collaboration.