<p>Underwater target detection is pivotal for marine resource exploration, environmental monitoring, and robotic applications. However, challenges such as low contrast, complex background interference, dense small targets, and multi-scale variations often lead to false or missed detections. To address these issues, this study proposes GAI-YOLO, an improved underwater object detection algorithm based on YOLO11n. First, a Global Edge Information Transfer Module (GEIT) is embedded into the backbone network. This module employs a Multi-Scale Edge Information Generator (MS-EIG) and an Edge Information Fusion Module (EIF) to efficiently propagate edge-enhanced features across the network, thereby strengthening feature representation and overall detection performance. Second, the Adaptive Receptive Field Enhancement (ARFE) module integrates dilated residual and re-parameterization techniques to dynamically adjust receptive fields, improving multi-scale target detection performance. Finally, Inner-WIoU is introduced as the loss function, which further improves small object localization accuracy and accelerates model convergence. Experiments on the DUO dataset demonstrate that GAI-YOLO achieves mAP@0.5, mAP@0.5:0.95, precision, and recall values of 86.0, 67.4, 87.5, and 76.8%, respectively. These results represent improvements of 1.6, 1.5, 1.9, and 0.6% over the baseline YOLO11n. Its generalization and robustness are further validated on the RUOD dataset, demonstrating the effectiveness of the proposed method for underwater detection tasks.</p>