<p>The quality of underwater images directly impacts the exploration and exploitation of marine resources. However, due to the complex underwater environments, underwater images often suffer from multiple distortions, such as blurriness, color cast, and low contrast. These distortions seriously degrade the performance of the downstream underwater vision tasks. While existing underwater image enhancement methods aim to improve visual quality, they often fail to preserve the critical features necessary for object detection. To address this issue, an enhancement method of underwater image with multi-distortion for object detection is proposed in this paper. Firstly, our method employs a bilateral constrained closed-loop adversarial enhancement module to overcome paired data scarcity. A detection-aware feedback module extracts gradient information to iteratively enhance detection-relevant features. Secondly, to handle multiple distortions, we propose a self-calibrating enhancement module to dynamically adapt to distortions at different spatial scales. Finally, a texture restoration module and a semantic-guided module are designed to jointly refine texture details and boundary integrity. Experimental results show that the proposed method significantly improves detection performance in multi-distortion underwater environments. Specifically, after enhancing the DUO dataset with the proposed method, YOLOv5 achieves an mAP of 60.4%. This corresponds to a 5.8% improvement over the raw images and a 4–13% increase compared to other advanced underwater image enhancement methods.</p>