Underwater object detection is extensively utilized in domains such as ocean exploration. Nonetheless, the intricate underwater environment results in challenges such as light attenuation and scattering, which diminish detection accuracy and do not satisfy the requisite standards. Therefore, this paper proposes an enhanced YOLOv8n-based model named UWD-YOLO to tackle these concerns. The dual-branch occlusion attention mechanism (DOAM) and dilated deformable convolutions module (DDCM) are the brains behind UWD-YOLO. In addition, the model calculates bounding box losses using the WIoU v3 method, which successfully handles regression issues associated with bounding boxes in both normal and exceptional scenarios. The enhanced UWD-YOLO model is trained and tested using the URPC2021 dataset to ensure its performance is completely validated. Results from experiments reveal that the UWD-YOLO achieves a mAP@0.5 of 79.3%, which is an improvement of 7% in detection accuracy (P) and 5.9% in precision (a mAP@0.5) over the YOLOv8n model. This approach is successful and has promising applications.