Speaker
Cristian Croitoru
(Technical University of Cluj Napoca)
Description
Faint satellite streaks are difficult to detect in astronomical images. High sensor dynamic range, stellar clutter, and atmospheric noise can hide these weak linear features. Global normalization can remove useful information because background noise can dominate the normalized image. We explore different preprocessing stages, from adaptive normalization to methods that enhance geometric and directional features. We then combine selected intensity and structural information into three-channel images for YOLOv8. This approach helps us examine which features improve detection across different observation conditions and confidence thresholds.
| Topic | Satellite Detection, Computer Vision |
|---|
Authors
Mr
Attila Fuzes
Cristian Croitoru
(Technical University of Cluj Napoca)