DCAS-EMAdp: Small Space Object Detection with Lightweight DCAS and Directional-Pooling Enhanced EMA

Abstract

Small Space Object Detection is a key component of Space Situational Awareness (SSA), which is essential for tracking and detecting fast-moving small space objects under complex environments. A major drawback is that detection models often cannot preserve the structural and contextual details of small or occluded objects. To address this problem, we propose a model named DCAS-EMAdp model, which combines a Dynamic Context-Aware Aggregation Strategy (DCAS) to dynamically expand the receptive field with an Efficient Multi-Scale Attention with Directional Pooling (EMAdp) module to enhance directional feature learning. The model is evaluated using real observational data obtained from the Geo-Informatics and Space Technology Development Agency (GISTDA), captured through ground-based telescope systems for space object monitoring. The experimental results show that DCAS-EMAdp achieves an mAP@50 of 91.4%. Furthermore, evaluations on mAP50@95, Precision, Recall, and F1-score confirm that DCAS-EMAdp has high capabilities in detecting small space object detection and occluded objects, and its performance is sufficient for practical applications in resource-constrained real-Time SSA systems.