TRANSFORMER-BASED GALAXY DETECTION SYSTEM IN THE SEARCH OF NEW GRAVITATIONAL LENS CANDIDATES

• 主要作者: Richard Sugandhi, Jonathan
• 格式: Final Project
• 語言: Indonesia
• 在線閱讀: https://digilib.itb.ac.id/gdl/view/76657

Gravitational lensing is an essential phenomenon in the development of astrophysics that significantly contributes to our understanding of the universe. However, direct observation of gravitational lensing phenomena is extremely challenging due to the vastness of the universe. In this study, deep learning approaches are discussed as one of the solutions to this problem. The focus of this research is on developing a galaxy detection system as one of the massive objects that have the potential to act as gravitational lenses. This study also aims to address issues in detection systems using heuristic approaches, which may fail to detect galaxies in certain cases and are unable to classify galaxy morphology. The galaxy detection system is built using the DEtection TRansformer (Carion, et al., 2020) architecture and trained on the dataset from the Sloan Digital Sky Survey (Margony, 1999). To obtain the best galaxy detection system, several experiments are conducted by varying the model's hyperparameters and training methods. The best experiment resulted in a mean average precision of 46.85 and an average precision at IoU threshold 0.5 (AP50) of 67.09. The existence of this deep learning- based galaxy detection system is expected to significantly contribute to the field of astrophysics by facilitating the detection and morphology classification of galaxies.