Enhancing Emergent Satellite Data for Non-Earth Imaging

The rapid evolution of commercial and government space systems is enabling emergent non-Earth imaging (NEI) capabilities for deep-space object characterization, optical space domain awareness (SDA), and other space surveillance applications. However, space-based imaging systems often operate under challenging observational conditions, including limited aperture size, restricted photon collection, low signal-to-noise ratio (SNR), sparse observation opportunities, spacecraft motion, and uncertain imaging geometry. These constraints can significantly degrade the quality and usability of collected data, limiting the performance of downstream detection, tracking, and characterization algorithms.

This work presents an end-to-end technical framework developed at Turion Space to improve the exploitation of degraded satellite observations for NEI applications. Rather than relying solely on hardware upgrades, the framework focuses on recovering and maximizing information already present in operational imagery. The approach combines three complementary processing components: (1) image enhancement and restoration to mitigate sensor and optical degradation, (2) image overlay and geometric validation to connect predicted object states with observed imagery, and (3) track-before-detect (TBD) processing to recover faint targets through temporal information integration.

The first component addresses image-level degradation introduced by the sensor and imaging environment. Raw observations are processed using astronomical calibration techniques, including bias subtraction, dark-current correction, fixed-pattern noise correction (FPN), non-uniformity correction (NUC), and flat-field normalization, to reduce detector-induced artifacts. Optical degradation is characterized through empirical point spread function (PSF) estimation using stacked stellar observations, enabling evaluation of restoration approaches including Wiener filtering, Richardson-Lucy deconvolution, Tikhonov regularization, and total variation regularization. Following restoration, image enhancement methods, including classical sharpening and learning-based super-resolution, are investigated to improve the visibility of weak structures and low-contrast targets. Because operational observations typically lack ground-truth reference images, performance is assessed using no-reference image quality metrics, visual inspection, and task-oriented evaluation, such as target detectability and localization accuracy.

The second component focuses on the geometric interpretation of satellite imagery. Reliable exploitation of optical observations requires accurate knowledge of spacecraft attitude, object ephemerides, and associated uncertainties. An image overlay pipeline is developed to project predicted resident space object (RSO) locations, covariance-based uncertainty regions, and catalog stars into the camera image frame. The pipeline incorporates orbital propagation from two-line element (TLE) data, state uncertainty estimation, spacecraft attitude information, and astronomical reference frame transformations. These overlays provide a direct comparison between predicted and observed locations, enabling evaluation of pointing accuracy, orbit prediction uncertainty, and target capture probability within the sensor field of view.

The third component addresses the challenge of detecting faint RSOs whose signals remain below conventional single-frame detection thresholds. TBD techniques are employed to exploit temporal coherence by integrating weak measurements along candidate object trajectories before making detection decisions. A hybrid TBD framework is investigated by combining Radon-transform-based motion proposal generation, particle-filter-based trajectory estimation, and generalized likelihood ratio test (GLRT)-based statistical evaluation. Candidate detections are ranked using a fused scoring approach that incorporates motion consistency, temporal persistence, photometric signal strength, and trajectory stability. This enables the recovery of weak object signatures that are difficult to identify from individual observations.

The proposed framework is evaluated using representative satellite imagery collected under challenging observational conditions. Experimental results demonstrate that detector calibration and image restoration improve the quality and consistency of observations used for downstream analysis. The image overlay pipeline provides quantitative validation between predicted orbital information and observed imagery by incorporating both target locations and uncertainty distributions. Furthermore, TBD processing demonstrates the ability to recover faint candidate objects through temporal integration beyond traditional frame-by-frame detection approaches.

This work demonstrates that integrated exploitation of satellite imagery, geometric information, and temporal observations can significantly improve the utility of emerging satellite data for NEI missions. By combining sensor characterization, uncertainty-aware data analysis, and advanced detection techniques, the proposed framework provides a pathway toward more capable space surveillance systems for future beyond-Earth-orbit missions, deep-space monitoring, and other challenging observation scenarios. The results also demonstrate the operational maturity of Turion Space’s NEI data processing stack for scalable commercial SDA workflows.