Strengthened Local Feature-Based Spatial–Temporal Tensor Model for Infrared Dim and Small Target Detection
Zehao Li, Shouyi Liao, Meiping Wu, Tong Zhao, Hongfeng Yu
Abstract
Infrared small and dim target detection is a difficult task in many fields, such as navigation, guidance, and early warning. The main challenges are inapparent target and complicated background. The low-rank and sparse model has been developed to address this problem, which provides a feasible and effective approach. In this article, the practicability of strengthened-local-feature-based spatial-temporal tensor (SLF-STT) model is considered to improve the optimization procedure of solution. A more consistent spatial–temporal tensor is designed, which helps sustain the function for non-Gaussian-like targets. A strengthened local feature map is acquired by the temporal-constrained Gaussian curvature filter and 3-D structure tensor (ST), which accelerates the convergence of iteration. Nonconvex approximation high-order singular value decomposition (HOSVD) model with adaptive total variation (TV) regularization is adopted, achieving better results of optimization. The proposed SLF-STT model can realize precise infrared small target detection with high efficiency, and it is robust for different scenes. A comprehensive test is carried out on open datasets to verify the performance of the proposed model, which shows its great advantages over other state-of-the-art (SOTA) methods.