X-shape Feature Expansion Network for Salient Object Detection in Optical Remote Sensing Images

Salient object detection in optical remote sensing images (RSI-SOD) is a valuable and challenging task. Some factors in RSI, such as the extreme complexity of scale and topological structure as well as the uncertainty of location of the salient object, significantly reduce the accuracy and completeness of salient object prediction. To address these issues, we propose a novel X-shape Feature Expansion Network (XFNet). Specifically, XFNet consists of a traditional encoder-decoder network, complemented by a new component called the X-shape Feature Expansion Module (XFEM). In XFEM, from the perspective of receptive field and multi-scale information, we utilize two branches to enhance the model's receptive field and multi-scale information. Moreover, we design a core component in XFEM to facilitate the fusion of feature of each branch. Extensive experiments conducted on two commonly used datasets demonstrate that our approach outperforms 11 state-of-the-art methods, including NSI-SOD and RSI-SOD methods.