Detection of slow-moving objects with LSST.

The Legacy Survey of Space and Time (LSST) will start running in 2025, spanning a decade, and will provide an unparalleled catalog of celestial objects observable from the Southern Hemisphere. Faint and distant objects will pose a challenge for the automated pipelines, and we want to improve our detection chances for these targets by using Machine Learning. To this purpose, we developed a binary classifier of slow-moving objects (SMOs) in LSST images by training a three-dimensional Convolutional Neural Network (CNN). As the survey has yet to commence, we have used simulated data, provided by the Dark Energy Science Collaboration, where we have injected simulated SMOs to create the training set. The simulated SMOs have been generated using the ephemeris of Trans-Neptunian objects and scaled to greater distances. We evaluated the network performance on an independent, smaller test set and found an accuracy of 90%. We are now investigating how the CNN performs as a function of the luminosity and apparent speed/orientation of the target SMOs.