Salient object detection (SOD) in optical remote sensing images (ORSI) has attracted considerable attention in recent years. With the rapid advancement of deep learning techniques, ORSI-SOD development has been remarkable. However, existing models continue to encounter significant challenges in processing certain scenarios, such as those consisting of low contrast, complex boundaries, and cluttered backgrounds. To address these challenges, we propose a progressive enhancement of the foreground feature network (PEFFNet) for ORSI-SOD, which is a novel three-stage design. In the first stage, a semantic-guided feature fusion module is introduced that adopts a top–down approach to effectively integrate multilevel feature information. This fusion strategy preserves the rich semantic information of the remote sensing object and accurately captures boundary detail features such that highly accurate initial optical remote sensing saliency map (ORSSM) can be generated. In the second stage, a simple and efficient feature enhancement module is designed, which consists of a background suppression module (BSM) and a bottom–up feature interaction module (BUFIM). The BSM utilizes an initial ORSSM to suppress background features, which significantly reduces interference from nonremote sensing regions. BUFIM enhances the feature representation of objects at different levels and optimizes object boundaries by fusing adjacent levels of features. In the third stage, a reverse attention decoding module is proposed to address pixel inhomogeneity and blurring in the remote sensing region. Experimental results demonstrate superior PEFFNet performance over other state-of-the-art models on three datasets on the basis of both quantitative and qualitative evaluations.