Detection of Streaks in Astronomical Images Using Machine Learning

Abstract

Satellites in low earth orbit (LEO) pose a challenge to astronomy observations requiring long exposure times or wide observation areas. As the number of satellites in LEO dramatically increases, it motivates an increased need for methods to filter out artifacts caused by satellites crossing into observation fields. This paper develops and evaluates a deep learning model based on U-Net to filter these artifacts from collected data. The proposed model is compared with two existing filtering methods on a dataset generated using the state-of-the-art tool Pyradon. Although the initial application of deep learning does include some unpredictable behavior not found in traditional algorithms, the proposed model outperforms the existing methods in overall accuracy while requiring significantly less computational time. This suggests that the application of deep learning to satellite artifact removal which has previously been underdeveloped in the literature may be an appropriate avenue.