Satellite Attitude Anomaly Detection in Ground Data Processing

In Earth observing satellite data processing, geometric image processing is the first and foremost step in generating accurate data products. The satellite’s attitude plays a crucial role in acquiring planned images, and any anomaly in attitude computation can lead to targeting inaccuracies and image distortion. Existing anomaly detection methods rely heavily on telemetry data and complex machine learning models, making them computationally intensive and less adaptable for real-time applications. In this work, we propose a novel graph-based RANSAC (Random Sample Consensus) algorithm to detect attitude anomalies in satellite profiles, ensuring early detection and correction in the ground data processing chain. Our method segments attitude profiles into structured components, identifies deviations from expected patterns, and estimates the maximum usable attitude range for longer image acquisitions. Unlike traditional approaches, the proposed method is computationally efficient, requires minimal prior knowledge of sensor-specific characteristics, and enables early anomaly flagging with improved reliability. Experimental validation on Cartosat-2S datasets demonstrates that the algorithm successfully identifies anomalous roll, pitch, and yaw profiles, providing insights into their impact on image quality. This approach enhances operational efficiency by reducing data rejection rates and enabling corrective measures before product generation.