A deep learning based method for the detection and classification of bright sources on Cassini ISS images

Recent advancements in artificial intelligence (AI) have opened new horizons for space exploration, especially in astrometry. In this work, we developed a deep learning-based algorithm to detect and classify bright sources—namely stars, satellites, and cosmic rays—in the Cassini Imaging Science Subsystem (ISS) images of the Saturn system. This novel approach leverages the precision of deep neural networks to process over 13,000 images, 1024×1024 pixels and with an exposure time <1 second, forming a robust dataset for training.

To generate labeled data without manual intervention, we designed a custom source detection algorithm using classical image processing techniques, such as mathematical morphology. Detected sources were then matched with star catalogs and ephemerides of Saturn’s moons to label stars and satellites; unmatched sources were classified as cosmic rays. The resulting dataset was used to train a YOLO (You Only Look Once) model—the state-of-the-art framework for detection and classification of objects on images and videos, that gained its popularity for its speed and accuracy. 

The network achieved strong classification results: cosmic rays were identified with 90% average precision and no false positives. Satellites were accurately classified 83% of the time, while stars proved more challenging due to their variability, achieving a 54% classification rate with 43% being misclassified as cosmic rays. 

Beyond detection, we used the classified data to study cosmic ray behavior in Saturn’s outer magnetosphere (15–100 Rs). Temporal variations were correlated with neutron monitor data from Earth, offering a broader view of cosmic ray activity in the solar system. Furthermore, analysis of the energy and directional characteristics of these particles demonstrates that ISS NAC images can be effectively repurposed for particle science. This AI-driven framework provides a new tool for exploring the Saturn system and could aid in the discovery of previously undetected moons or energetic events.