Underwater object detection plays an important role in marine environment monitoring, resource exploration, ecological protection, and underwater operations. However, the complex underwater environment, together with the absorption and scattering of light, often leads to color distortion and blurred details in images; in addition, underwater targets are typically small, densely distributed, and prone to occlusion, which further increases the detection difficulty. To address these challenges, this paper proposes an improved underwater object detection algorithm, termed REL-YOLO, based on YOLOv11. First, to cope with complex backgrounds and severe noise in underwater environments, Receptive Field Attention (RFA) and Coordinate Attention (CA) are integrated to enhance feature extraction and discrimination capability. Second, an Edge Information Enhancement Module (EIEM) is designed to strengthen contour features of targets, thereby improving the detection sensitivity to blurred edges and small objects. Finally, a Multi-Scale Large Kernel Attention (MSLKA) module is constructed to enhance the perception of multi-scale and occluded targets through large-kernel convolutions at different scales. Experimental results show that the proposed method achieves detection accuracies of 86.5% and 87.7% on the public URPC2020 and RUOD datasets, representing improvements of 1.8% and 1.0% over YOLOv11. Compared with other methods, REL-YOLO demonstrates higher accuracy, better generalization, and stronger robustness in multi-scale underwater target detection.