Topological Prior-Embedded Spatial–Temporal Tensor Approximation Model for Infrared Small-Target Detection

Infrared small-target detection is crucial in both military and civilian applications, yet it remains challenging to distinguish true targets from complex backgrounds and target-like flickering noise. To address this issue, a novel topological prior-embedded spatial-temporal tensor approximation (TP-STTA) model is proposed. The framework first design multi-topological feature extraction (MTFE) model, in which area, diagonal, and volume attributes are explored, which effectively characterize the spatial properties of small targets and suppress background clutter and flickering noise. These features not only provide an auxiliary detection response but, more importantly, serve as target-aware priors to guide the subsequent tensor-based optimization. Building upon this, a multi-prior tensor approximation (MPTA) model jointly exploits spatial-temporal correlations by integrating the multi-attribute topological priors from MTFE with background-oriented smoothness and low-rank constraints, leading to stable and robust target enhancement. Finally, the responses from MTFE and MPTA branches are fused, and temporal trajectory consistency is exploited to further suppress false alarms. Experimental results on six real infrared sequences demonstrate that the proposed TP-STTA achieves accurate and reliable small-target detection with significant background suppression.