Salient object detection in optical remote sensing imagery (ORSI-SOD) presents significant challenges due to complex backgrounds, low-contrast and ambiguous object boundaries, and substantial scale variations among targets. To address these issues, this paper proposes a novel deep learning architecture, termed EMANet, which integrates three key modules: an Edge-Enhanced Feature Module (EEFM) to improve boundary localization; a Multi-Scale Edge Modulation (MSEM) module for scale-aware feature refinement; and an Attention-Guided Feedback Enhancement Module (AGFEM) to enhance semantic consistency across network layers. The proposed framework leverages multi-scale feature fusion, edge-guided supervision, and attention-driven feedback to generate high-precision saliency maps under varying object scales and complex scene conditions. Extensive experiments on two publicly available datasets, ORSSD and EORSSD, demonstrate that EMANet outperforms state-of-the-art methods in terms of S-measure, F-measure, and mean absolute error (MAE). Furthermore, ablation studies validate the individual and combined effectiveness of each module. These results confirm that EMANet offers a robust and efficient solution for ORSI-SOD, particularly in detecting small, irregular, and low-contrast targets in challenging remote sensing scenarios.