Infrared Small Target Detection via Schatten Capped pNorm-Based Non-Convex Tensor Low-Rank Approximation

For infrared (IR) small target detection, we propose a new spatial-temporal tensor (STT) decomposition model based on the tensor Schatten capped $p$ norm (TSC $p\text{N}$ ) and total variation (TV) regularization. First, to explore spatial and temporal information, we construct an STT and introduce the prior weight map. Then, replacing the nuclear norm used to define tensor nuclear norm (TNN) with the nonconvex Schatten capped $p$ (SC $p$ ) norm, we propose the TSC $p\text{N}$ to approximate tensor rank and thus recover the low-rank background tensor. Next, to effectively eliminate background clutter from the sparse component, TV regularization is applied to constrain the sparse term. Finally, the proposed model is solved by the alternating direction method of multipliers (ADMM). Experimental results demonstrate the effectiveness and superiority of our proposed method in detecting IR small targets from various challenging sequences.