Research on Improved Detection Method of Orbital Foreign Bodies in YOLOv10

Aiming at the issues of low accuracy, low efficiency and misdetection existing in current orbital foreign object detection methods, an enhanced YOLOv10 orbital foreign object detection approach was proposed, and a lightweight model for the improved YOLOv10 orbital foreign object detection was constructed. Firstly, the EMA attention mechanism was incorporated into the C2f and C2fCIB modules of the YOLOv10 backbone network to enhance the capture of spatial information at different scales and strengthen the model's ability to express multi-scale features. Finally, cross-scale horizontal jump connection combined with weighted feature fusion was introduced in the neck layer to optimize the feature fusion structure and reinforce the multi-scale feature fusion capability. The proposed model's [email protected] can reach 97.4%, which is 3.1% higher than that of YOLOv10. Additionally, the frame rate, parameter number, computational effort and model volume of the model detection are also superior to those of most mainstream target detection algorithms. The experimental results indicate that the proposed model can effectively enhance the detection accuracy of orbital foreign bodies.