<p>RGB-D salient object detection aims to locate salient targets by jointly leveraging RGB and depth information. Most existing methods progressively fuse the corresponding semantic features from RGB and depth modalities across multiple layers. However, the semantic information contained in RGB and depth images is inherently different, and direct fusion may lead to feature redundancy or modality conflicts. To address this issue, we adopt an RGB-dominated and depth-assisted feature interaction strategy, where the highest- and lowest-level RGB and depth features are fused to generate a coarse guidance map that assists the subsequent RGB feature enhancement process. Therefore, we propose a coarse feature generation and multi-scale pooling feature fusion module to provide guidance. In addition, we design a multi-scale depth convolutional feature enhancement module, which is combined with a progressive decoding strategy to generate the final refined prediction map. Finally, all the proposed modules are integrated into a unified framework, named CGMSNet. Extensive experiments on 6 benchmark datasets demonstrate that our method has achieved the state-of-the-art performance with high efficiency in RGB-D salient object detection.</p>