Hybrid deep learning approach for detecting and classifying space debris in low Earth orbit

The growing concentration of debris in space poses dangers to the operation of satellites, spacecraft, and the sustainability of future space activity in general. Hypervelocity impacts by space debris may impair essential onboard mechanisms, damage costly infrastructure, and create new orbital debris. In response, this paper presents a novel deep learning-based system architecture for space debris detection and classification within the region of low-Earth orbit (LEO). As regards debris detection, both YOLOv8 and YOLOv12 models augmented with the Convolutional Block Attention Module (CBAM) were used in this study due to their capability of detecting tiny and occluded space debris objects. Our experiments demonstrate that YOLOv8 + CBAM yielded mAP@50 of 0.6718 and recall of 0.6707 on SPARK data, whereas YOLOv12 + CBAM had the best recall on the Space Debris dataset.For debris classification, this paper introduces DebriXNet, a lightweight convolutional neural network specifically designed for resource-constrained onboard environments. DebriXNet contains approximately 1.6 million parameters and achieved an average multi-class classification accuracy of 84.65% on the SPARK dataset while maintaining low inference latency suitable for real-time deployment on embedded spacecraft platforms. The proposed hybrid framework combines attention-enhanced debris detection with efficient lightweight classification to support intelligent onboard decision-making, collision avoidance, and space situational awareness (SSA). Overall, the proposed system provides a practical and scalable solution for autonomous debris monitoring and contributes toward improving the safety, efficiency, and sustainability of future space missions.