Infrared small target detection based on clustering and tracking

Infrared (IR) small target detection has been widely used in civilian and military applications. Although low-rank and sparse tensor decomposition theory has been widely employed, the estimations of target and background are still not precise enough. This paper proposes an IR small target detection method based on improved clustering and Bayesian guided-tracking regularization (STD-ICBT). Specifically, a 3-D spatial-temporal tensor is constructed first. Secondly, we improve the K-means clustering algorithm for lower-rank background fiber clusters and design an improved K-means clustering-based background estimation method, making it more accurate for background estimation. Furthermore, we design an efficient ADMM-based optimization algorithm for solving the target detection model. Compared with six state-of-the-art competitive methods, it demonstrates the superiority of STD-ICBT in terms of target detectability (TD), background suppressibility (BS), and overall performance