Underwater image segmentation is currently a challenging issue in underwater image technology and has gradually attracted more attention from researchers. Despite significant advancements in this field, the complex underwater environment continues to pose difficulties, leaving room for improvement in segmentation accuracy. To enhance underwater image segmentation performance, the Multi-Scale Feature Fusion Module (MSFFM) proposed in this paper is designed for upsampling feature fusion. It strengthens the incorporation of shallow semantic information. Additionally, the feature fusion process incorporates Channel prior Convolutional attention(CPCA). Finally, the optimization process utilizes a composite loss function incorporating Dice Loss and Focal Loss. Results of experiments carried out on the SUIM dataset demonstrate that the model we proposed achieves excellent segmentation accuracy for underwater imagery.