Infrared Dim and Small Target Detection Based on Robust CUR Decomposition

Infrared dim and small target detection algorithms play a crucial role in military applications. To address the issues of high computational complexity, poor interpretability, and insufficient utilization of image information—which often result in inefficient model convergence and incomplete target retention—this paper proposes a novel robust Column-Row (CUR)-based matrix decomposition method. First, we construct a novel sampling strategy to reconstruct the original image through analysis of target and background characteristics, establishing a foundation for matrix decomposition. Subsequently, based on the reconstructed images, we propose a new CUR-based matrix solution method to solve the low-rank terms in the infrared dim target detection model, effectively reducing computational complexity. Finally, we employ Randomized Block Krylov Methods as a replacement for traditional Singular Value Decomposition (SVD) to further enhance computational efficiency. The visual detection results of each scene and various evaluation indexes such as Receiver Operating Characteristic (ROC) curve show that the proposed method has satisfactory computational efficiency and detection performance.