YOLO-Dynamic: A Detection Algorithm for Spaceborne Dynamic Objects

Abstract

Ground-based detection of spaceborne dynamic objects, such as near-Earth asteroids and space debris, is essential for ensuring the safety of space operations. This paper presents YOLO-Dynamic, a novel detection algorithm aimed at addressing the limitations of existing models, particularly in complex environments and small-object detection. The proposed algorithm introduces two newly designed modules: the SC_Block_C2f and the LASF_Neck. SC_Block_C2f, developed in this study, integrates StarNet and Convolutional Gated Linear Unit (CGLU) operations, improving small-object recognition and feature extraction. Meanwhile, LASF_Neck employs a lightweight multi-scale architecture for optimized feature fusion and faster detection. The YOLO-Dynamic algorithm’s performance was validated on real-world images captured at Antarctic observatory sites. Compared to the baseline YOLOv8s model, YOLO-Dynamic achieved a 7% increase in [email protected] and a 10.3% improvement in [email protected]:0.95. Additionally, the number of parameters was reduced by 1.48 M, and floating-point operations decreased by 3.8 G. These results confirm that YOLO-Dynamic not only delivers superior detection accuracy but also maintains computational efficiency, making it well suited for real-world applications requiring reliable and efficient spaceborne object detection.

Keywords

• spaceborne dynamic object detection
• YOLOv8
• SC_Block_C2f
• LASF_Neck
• multi-scale feature fusion