To solve the detection problems caused by common blurring distortion, large differences in target scale, and mutual occlusion in underwater images, a new underwater image object detection method DBSD-YOLOv8 based on YOLOv8n model improvement is proposed. Introducing DICAM as a pre enhancement module for feature extraction in the backbone structure to alleviate the common problems of scale degradation and uneven light attenuation in underwater images; Adopting the dual phase feature aggregation module BFAM as an alternative structure to the C2f module to maintain high-precision recognition in complex scenarios such as noise interference and boundary blurring. Embedding the separation enhanced attention mechanism SEAM in the Neck layer enhances the model"s feature modeling ability in occluded and blurry environments. The detection head adopts DyHead detection head, which utilizes its multi-scale and spatial position perception ability to improve the detection accuracy of targets of different sizes and positions. The experimental results show that the improved model performs well on the RUOD dataset mAP@0.5 and mAP@0.5:0.95 achieved 0.862 and 0.622, respectively, which were 3% and 3.3% higher than the original YOLOv8n model.