Instance Segmentation of Messier Objects: YOLO vs. Mask R-CNN

Artificial intelligence has generated revolutionary advancements in the realm of Astronomy. One of the domains where AI can have a significant impact is machine vision, wherein the captured images from telescopes and satellites can be examined. However, detecting objects such as planets, comets, asteroids, galaxies, etc., in celestial imagery is a challenging task due to the existence of intense light sources, noises, and low contrast, and it requires multiple advanced preprocessing techniques prior to feeding the data into deep learning algorithms. In this paper, our objective is to conduct image segmentation on eleven Messier objects. The Messier objects constitute a collection of cosmic objects documented by the French astronomer Charles Messier during the ۱۸th century. This compilation, recognized as the Messier Catalog, encompasses some of the most famous and visually captivating celestial entities observable from the Northern Hemisphere of Earth. We used Mask R-CNN with backbones of ResNet۵۰ and ResNet۱۰۱, and also YOLOv۸ for instance segmentation and reached the best precision of ۹۴%. The results substantiate the effectiveness of our model.

کلیدواژه ها:

celestial objects ، Telescope ، Yolov۸ ، constellation ، convolutional neural network ، deep learning