Space Micromotion Targets Recognition With Denoising and Spatial-Temporal Attention Network

The recognition of space micro-motion targets plays a critical role in modern aerospace surveillance and defense systems. However, noise interference and insufficient feature representation in conventional methods limit recognition performance. To address these challenges, this paper proposes a Denoising and Spatial-Temporal Attention Network (DSTA-Net) for space micro-motion target recognition using a Dynamic Range-Doppler Sequence representation. The Dynamic Range-Doppler Sequence integrates time, micro-range, micro-Doppler, and radar cross-section (RCS) attributes into a unified structure, enabling a comprehensive depiction of target signatures. A multi-scale denoising module is first applied to suppress noise while preserving scattering centers. The cleaned sequences are then fed into a recognition network that incorporates a temporal attention unit, comprising both dynamic and static branches, as well as a multi-scale coordinate attention unit, which locates scattering centers. Experimental results demonstrate that the proposed approach achieves robust recognition performance across 12 categories of targets, especially under low signal-to-noise conditions.