Abstract
Infrared small target detection is important for remote sensing surveillance, precision guidance, and early warning, but reliable detection in complex scenes remains difficult because small targets usually occupy only a few pixels and exhibit weak contrast and low signal-to-noise ratio. Complex background clutter, including clouds, sea waves, ground edges, and sensor noise, can easily cause missed detections and false alarms. This paper presents a complementary feature enhancement (CFE) framework for infrared small target detection in complex scenes. The proposed method formulates the input image sequence as a spatial–temporal tensor decomposition problem and jointly models three complementary priors: adaptive background representation, asymmetric spatial-temporal consistency constraint, and saliency-guided sparse target enhancement. Specifically, an adaptive nonconvex background representation is designed to assign nonuniform weights to singular values, thereby improving structured background estimation. An asymmetric spatial-temporal consistency constraint imposes different smoothness strengths on spatial and temporal dimensions, which improves robustness to dynamic background variations. To enhance weak targets, local contrast and directional saliency cues are integrated to construct a saliency prior, which is embedded into the sparse target component rather than used as an independent detector. Experiments on representative infrared sequences show that the proposed complementary feature enhancement framework enhances dim target responses, suppresses residual clutter, and achieves robust detection under challenging backgrounds.