Stack-CNN algorithm: A new approach for the detection of space objects

We present a new trigger algorithm combining a stacking procedure and a Convolutional Neural Network that could be applied to any space object moving linearly or with a known trajectory in the field of view of a telescope. This includes the detection of high velocity fragmentation debris in orbit. A possible implementation is as trigger system for an orbiting Space Debris remediation system. The algorithm was initially developed as offline system for the Multiwavelength Imaging New Instrument for the Extreme Universe Space Observatory (Mini-EUSO), on the International Space Station. We evaluated the performance of the algorithm on simulated data and compared it with those obtained by means of a more conventional trigger algorithm. Results indicate that this method would allow to recognise signals with 1% Signal over Background Ratio (SBR) on poissonian random fluctuations with a negligible fake trigger rate. Such promising results lead us to not only consider this technique as an online trigger system, but also as an offline method for searching moving signals and their characteristics (speed and direction). More generally, any kind of telescope (on the ground or in space) such as those used for space debris, meteors monitoring or cosmic ray science, could benefit from this automatized technique. The content of this paper is part of the recent Italian patent proposal submitted by the authors (patent application number: 102021000009845).