<p>Accurate detection and timing of impulsive maneuvers by resident space objects is critical for ensuring space domain awareness, collision avoidance, and overall mission assurance. Conventional innovation-based methods that rely on full-state filters degrade in performance within the highly nonlinear dynamics of the Circular Restricted Three-Body Problem, particularly when maneuvers are executed in ways that render them near “invisible” to observer spacecraft. To address this challenge, this paper introduces a data-driven framework that leverages modal decomposition of measurement innovation histories to enhance maneuver detection. Within this framework, an Extended Kalman Filter is employed to propagate and update state estimates. Simultaneously, maneuver identification is achieved by detecting sharp spikes in the normalized innovation-squared history once projected onto carefully selected reduced-order modal subspaces. A suite of adversarial and strategically designed maneuvers are shown and tested to evaluate the capability of the reduced-order modal innovation against traditional full-spectrum identification methods. Results demonstrate that modal decomposition can reliably isolate maneuver signatures even in scenarios where conventional innovation tests fail. Finally, an adaptive Extended Kalman Filter architecture is proposed in which the state covariance is inflated upon maneuver detection, thereby enhancing robustness and improving subsequent estimation accuracy and robustness.</p>