Infrared Patch-tensor Model with A New Tensor Nuclear Norm for Small Target Detection in the Ground Background

Abstract

Accurate detection of infrared small target in complex scenes plays a vital role in many practical applications. In the infrared patch-tensor (IPT) model, small target detection is regarded as a convex optimization problem of tensor robust principal component analysis (TRPCA), which separates low tubal rank and sparse tensors. Tensor singular value thresholding is employd as the proximal operation of tensor nuclear norm to estimate the low-rank background components. Considering the defect that the local structural weights in the Reweighted IPT shrink excessively small targets, an effective local prior map is used to better mine target and background information. The local prior and sparse enhancement weight are combined to obtain the final adaptive weight. Alternating direction method of multipliers (ADMM) is utilized to solve the model. Extensive experiments of infrared sequence images based on ground background prove that our algorithm has significant advantages in real-time and detection capability compared with state-of-the-arts.