TV regularized low-rank sparse decomposition of infrared polarization component image for small target detection under complex backgrounds

Abstract

Infrared small-target detection in complex backgrounds remains a significant challenge. Infrared polarization imaging provides an effective means of enhancing target contrast. However, edge clutter from the observed scene and polarization resolution noise greatly restrict the application of infrared polarization images for small target detection, and many typical and contemporary infrared target detection algorithms are unsuitable for such images. To address these challenges, we construct an image representing the infrared polarization component using the partial elements of the Stokes vector. A total variation (TV) regularized low-rank sparse decomposition method is proposed to decompose infrared polarization component images into target, noise, and background. Owing to the target’s geometric regularity and the discrete nature of noise, TV regularization is employed to suppress impulse noise and edge clutter that resembles target structures, resulting in a denoised and more accurate target representation. The augmented Lagrange multiplier method is developed to solve the proposed optimization problem. An experiment was conducted on the detection of a hovering unmanned aerial vehicle target against a complex background. Compared with state-of-the-art infrared small-target detection algorithms, our method achieves highly effective detection for both low-contrast and high-contrast targets.