Infrared small target detection based on multiscale low-rankness and firm thresholding function

Abstract

Recently, infrared (IR) small target detection based on low-rank and sparse decomposition (LRSD) has gained increasing attention among civilian and military fields. Whereas, the LRSD-based methods face the challenges of accurately mathematically describing the intrinsic characteristics of each component and effectively separating targets from non-targets. This paper proposes a target detection model with multiscale low-rankness and firm thresholding function (MSR-FT). Specifically, a high-pass filtering spatial–temporal tensor corresponding to the original IR sequence is constructed. Secondly, we define a firm thresholding function-based norm for target characterization, leading to more accurate sparse target estimation. Meanwhile, using a multiscale low-rank background decomposition technique, we introduce a multiscale Log-based tensor nuclear norm, which ensures that the estimated background fully considers the global and local low-rank properties from different scales. Furthermore, with a posterior information feedback strategy, we propose a target detection method called MSR-FT. Through an optimization scheme based on the alternating direction method of multipliers (ADMM), it proves that MSR-FT exceeds seventeen competitive IR small target detection methods on six IR sequences from the perspectives of target detectability (TD), background suppressibility (BS), and overall performance.