ML-based Algorithm for Streak Detection in Astronomical Images

Recent years have proven that the non-stop rise in the number of Resident Space Objects (RSOs) represents a serious threat to space sustainability. As a consequence, Space Surveillance & Tracking (SST) is becoming a key field since its purpose is to detect and track RSOs. This objective can be achieved with several solutions including optical sensors, among which ground telescopes. The discovery of new objects and the surveillance of the existing ones is made possible by the survey mode of such sensors, characterized by images taken with large exposure times where RSOs appear as strips of pixels (i.e., streaks). This work addresses the streak detection problem in survey images by automatically isolating such strips by means of a Convolutional Neural Network (CNN), whose training dataset's annotations are provided by an image processing pipeline applied to the raw astronomical frames.