Infrared Small Target Detection via Nonconvex Tensor Fibered Rank Approximation

Abstract

Infrared small target detection plays an important role in precision guidance, infrared warning, and other applications. The infrared patch-tensor (IPT) model has good detection performance, but some challenges still exist, such as the inaccurate representation of the background rank and poor robustness against noise and sparse interference. In order to solve these problems, a new IPT model is proposed in this article. First, to approximate the tensor rank more reasonably, t-SVD is generalized to multimodal t-SVD, and the tensor fibered rank is introduced. Moreover, the tensor fibered nuclear norm based on the Log operator (LogTFNN) is used to nonconvex approximate tensor fibered rank. Second, to suppress sparse interference such as strong edges and corner points, the prior information is extracted by the local structure tensor. Third, the hypertotal variation (HTV) is used as a joint regularization term to remove noise. Then, the alternating direction method of multipliers (ADMM) is used to solve the model. The proposed algorithm was tested on the 20 single-frame infrared images and six sequences of real scenes. Lots of experiments demonstrate that this algorithm has the robustness to noise and different scenes. Different evaluation metrics also show that the proposed algorithm has a significant superiority in detection performance compared with various state-of-the-art methods.