The remote sensing image can be effectively utilized for rapid sea fog detection. However, directly applying the semantic segmentation model of close range images to remote sensing images is challenging due to significant disparities between the two types of imagery. Therefore, we propose the SF-SegNeXt model based on the SegNeXt architecture, which incorporates a channel attention mechanism to enhance feature extraction capabilities. Additionally, we introduce a Foreground-Semantic decoder and a corresponding loss function to improve the model’s ability in detecting foreground objects within remote sensing images. To evaluate our proposed model and advance research in sea fog detection, we construct a dedicated sea fog dataset and conduct experiments accordingly. Results demonstrate that our SF-SegNeXt model outperforms existing approaches by achieving superior performance in foreground object detection within remote sensing images while enabling high-precision sea fog identification.