Automatic Astrophotography Editing Method Using Deep Learning

Constellation detection in astrophotography is a difficult task because the stars appear very different depending on the photographer and the equipment, and because there are many similar alignments of stars. Although there are some previous studies on constellation detection, they either require a long time to totally match the bright stars in the astrophotography with the major stars that make up each constellation, or they match constellations that are known in advance in the astrophotograph, which is not very practical. Therefore, we propose a method using deep learning that can detect constellations more efficiently than total matching, even when the constellations in the astrophotogra-phy are not known in advance. Specifically, the object detection network that has been trained on the characteristics of each constellation by original data sets detects the rough location of each constellation, and from there, fine point matching is used to accurately detect the location of the constellation. As a result, the proposed method can detect constellations in the test images with 80 percent accuracy, and the average detection time was approximately 1 seconds in CPU time. This confirms that the proposed method is accurate and efficient. In addition, based on this method, we aim to realize automatic editing for astrophotography, such as complementing stars that are not in the astrophotography and adding color information to stars.