Abstract: As the number of mega-constellations increases, maneuver detection of non-cooperative satellites becomes an important part of Space Situational Awareness (SSA). The nonsingular thrust-Fourier coefficients (NSTFC) model can effectively fit the maneuvers of mega-constellations, but the feasibility of the online maneuver detection method based on this model needs to be verified. The method uses the orbital variations estimated by a filter as the classification attributes and uses the Naive Bayes as the classifier. The calculation results based on simulation show that this method can effectively identify three classes of space events: no maneuver, orbit raising, and orbit descending. The Macro F_1 score of classification can reach as high as 97.1 \% . The verification shows that this method may help improve the efficiency and accuracy of maneuver detection in the cataloging.