Due to the poor quality of underwater images, they are usually not accurate when used for object detection. Most of the currently available solutions involve pre-processing, such as underwater image enhancement, to improve object detection accuracy. However, the pre-processing approach is to improve the subjective perception of the human eye, which does not necessarily improve the object detection performance and also consumes a lot of computational resources. Therefore, in this paper, we innovatively combine these two tasks and propose Shape-Guided Detection Network (SG Detection). We fuse underwater image enhancement and object detection into one network, where object detection can share clean features through joint learning, thus facilitating underwater object detection , and the object detection results will also guide the direction of underwater image enhancement. Meanwhile, in the SG Detection network, we embed the object shape prior features as a learnable module, and design the Shape Priori Enhancement module to achieve the full fusion and interaction between the prior features and local features. In addition, we innovatively adopt a combination of explicit and implicit constraint approach, which ultimately achieves high object detection accuracy. We have conducted extensive experiments on public available datasets, and we find that our approach achieves up to 10% mAP improvement in object detection performance and significantly increases the processing speed.