Image-Based Transient Detection Algorithm for Gravitational-Wave Optical Transient Observer (GOTO) Sky Survey

The paper proposes an alternative method to detect transients for an optical telescope using a deep learning algorithm. While the previous studies followed the conventional method, a classification, focusing on the manually extracted features of transients, the alternative method does the detection focusing on imagery instead. The algorithm is based on a famous UNET model which can do both classification and segmentation at the same time. Initial setups are used to test the capability of the model. In the same way, some data is fused with noise to see the model limitation. The result is a map showing where the objects are located, identified by binary class numbers. Both results either with or without noise are provided. This includes a comparison between different batch size setups as one of the key parameters for deep learning. As a preliminary to further studies, a list of possible parameters is also given.