<p>Existing methods for salient object detection (SOD) in low-light images primarily rely on feature extraction in the spatial domain, which is insufficient for capturing effective information in complex environments. Although multi-level feature fusion provides some help, it is still challenging to generate accurate saliency maps. To tackle the aforementioned issues, we propose a novel spatial-frequency domain aware network (SFANet) consisting of four components: a spatial-frequency domain feature extractor (SFDFE), a boundary-aware module (BAM), a localization enhancement module (PEM), and a feature fusion attention module (FFAM). Specifically, we design the SFDFE to fully extract the texture information and intrinsic structure of the image. The detailed features of the extracted boundaries are enriched using the full interaction of multi-scale information realized through the BAM. In addition, a PEM is used to achieve precise target localization. Finally, the FFAM serves to merge entirely the different scale information to achieve accurate object detection. We validate the effectiveness of SFANet on three datasets and show that it substantially surpasses existing state-of-the-art SOD models both qualitatively and quantitatively. Furthermore, we explore the potential of SFANet for other vision tasks that are closely related to SOD, including SOD in natural scenes, video SOD in low-light scenes, and camouflaged object detection. Experimental results confirm strong generalization capability and robust performance of SFANet.</p>