Under low data rates, the observational data is sparse, and the noise have a significant impact. Under such circumstance, filtering algorithms tend to rely more heavily on predictions. However, the absence of maneuver information usually results in significant model errors, ultimately affecting the accuracy of states estimation and potentially leading to filter divergence. To address this issue, we have constructed a model set specific to satellite maneuvers and proposed an Improved Interacting Multiple Model (IMM) algorithm based on a designed maneuver detection model and estimation model. We have developed a maneuver estimation model to estimate the maneuvers, and to perform pruning on the maneuver models before estimation. The algorithm dynamically controls the introduction of maneuver models via the detection model and estimation model. When a maneuver is detected, the algorithm introduces the maneuver model set for fusion estimation; otherwise, it degrades into a Kalman filter based on a non-maneuver model. Simulation results indicate that the proposed algorithm effectively tracks targets in low data rate scenarios while maintaining high estimation accuracy.