Autonomous navigation of unmanned surface vessels (USVs) relies on the semantic segmentation of images from onboard cameras, but the current segmentation accuracy between categories still needs to be improved. Due to the significant differences between maritime scenes and land environments, traditional semantic segmentation methods do not perform well at sea, and the existing maritime semantic segmentation techniques also mainly focus on utilizing the water shoreline edge features to guide semantic segmentation while ignoring the edge features of other categories. In order to solve the above problems, this paper designs a semantic segmentation network based on the reinforcement of edge features for each category in a maritime scenario. The network extracts edge features of each category by introducing the Edge Stripping Module (ESM). It enhances the edge information in the semantic feature fusion process by the Edge Reinforcement Block (ERB), significantly improving the segmentation accuracy. The network was experimentally evaluated on the MODD2 dataset using the MODS evaluation system and compared with methods such as PSPNet, Deeplabv3+, and WaSRNet, and the results verified the effectiveness of the network in this paper.