针对高轨卫星自主无源探测预警任务中对袭目标轨道机动行为的检测发现时间滞后大、准确率低等问题，提出了一种基于双注意力对比学习(Dual Attention Contrastive Representation Learning for Time Series Anomaly Detection, DCdetector)的轨道机动智能检测方法。首先，为了提升无源测量条件下的目标轨道状态可观测性，在惯性系下建立了高阶受摄的非线性轨道动力学模型和视线角测量模型，共同构成无源探测相对导航系统。然后，设计了基于平方根容积卡尔曼滤波器的相对导航估计算法，并构建了反映状态估计偏差的归一化新息，用于表征目标机动行为。接着，设计了以归一化新息序列作为输入的DCdetector无监督神经网络机动检测模型，并通过对比学习训练获得了用于目标轨道机动判决的阈值，最终根据归一化新息的异常分数是否超过该阈值来完成机动判决。最后，以典型的GEO轨道预警任务场景为例对所提方法进行了数值仿真实验验证与性能分析。仿真结果表明，在测角不确定性为0.005°时，所提算法能以超过90%的成功率检测出350km距离目标不低于0.5m·s–1的小量级脉冲机动，且平均的机动发现时间延迟小于360s。通过与现有三种检测方法对比表明，本文方法在检测成功率、稳定性和时间误差方面均表现最优，显著提升了高轨目标轨道机动无源检测的综合性能。

To address the problems of large detection delays and low accuracy in detecting orbital maneuvers of incoming targets in autonomous passive detection and early warning missions for high-orbit satellites, this paper proposes an intelligent detection method for orbital maneuvers based on Dual Attention Contrastive Representation Learning for Time Series Anomaly Detection (DCdetector). First, to improve the observability of target orbital states under passive measurement conditions, a high-order perturbed nonlinear orbital dynamics model and a line-of-sight angle measurement model are established in the inertial frame, which together form a passive detection relative navigation system. Then, a relative navigation estimation algorithm based on the square-root cubature Kalman filter is designed, and a normalized innovation reflecting the state estimation bias is constructed to characterize the target maneuver behavior. Next, an unsupervised DCdetector neural network maneuver detection model is designed using the normalized innovation sequence as input. A threshold for orbital maneuver judgment is obtained through contrastive learning training, and maneuver detection is ultimately performed by checking whether the anomaly score of the normalized innovation exceeds this threshold. Finally, numerical simulation experiments and performance analysis are conducted to validate the proposed method using a typical GEO orbit early warning mission scenario. Simulation results show that, with an angle measurement uncertainty of 0.005°, the proposed algorithm can detect small-magnitude pulse maneuvers of 0.5m·s–1 or greater for a target at a distance of 350km with a success rate exceeding 90%, and the average maneuver detection time delay is less than 360s. Comparison with three existing detection methods demonstrates that the proposed method achieves the best performance in terms of detection success rate, stability, and temporal error, significantly improving the overall passive detection performance of orbital maneuvers for high-orbit targets.