基于双域协同感知的太空红外小目标检测方法OA

With the increasing frequency of space activities,space target detection has become increasingly important for orbital space safety.small dim infrared targets in space scenes are difficult to detect because of their weak characteristics and susceptibility to complex background interference.To address the loss of small-target information and confusion with similar backgrounds in existing methods,a space-based infrared small-target detection approach based on dual-domain collaborative perception is proposed.In the backbone network,a spatial-frequency synergistic convolution module is designed,which integrates learnable center-surround differences with adaptive spectral filtering to enhance the feature extraction for dim and small targets and achieve collaborative modeling of local details and global structures.Forthermore,a hierarchical attention aware is inserted between the neck and the detection head,where multi-scale hierarchical attention is employed to promote inter-layer feature interaction,thereby improving the discriminability between targets and background.Experiments conducted on a self-constructed space infrared dim and small target dataset demonstrate that the proposed method achieves a precision of 97.82%,a recall of 94.63%,and an F1-score of 96.02%,outperforming mainstream methods across all metrics and effectively improving detection performance under complex space backgrounds.