The characterization of satellite behavior is of paramount importance in Space Surveillance Awareness. One crucial aspect of studying satellite behavior involves the detection of out-of-station-keeping maneuvers within satellite trajectories. Since each satellite's station-keeping behavior is unique, extracting relevant information from historical data is crucial for modeling the Pattern of Life. In this context, we propose integrating a state-of-the-art data mining technique, namely the Matrix Profile, into a neural network architecture. Our approach highlights the connection with non-local models. The proposed model exhibits promising results for detecting out-of-stationkeeping maneuvers in Low Earth Orbit satellite trajectories, showcasing its effectiveness in addressing the challenges of unbalanced data.