In underwater hull cleaning tasks, the detection and precise identification of fouling and obstacles by target detection technology are critical for enhancing the efficiency of automated cleaning. The current underwater-hull cleaning robots primarily rely on manual visual inspection for environmental perception and target localization, which imposes efficiency bottlenecks and safety hazards. An improved model, YOLO-HC, is proposed based on the YOLOv11 framework. The model enhances cross-scale feature extraction capability by constructing a Multi-scale Dilated Attention module (MSDA_C2PSA), optimizes multi-level target representation fusion by using a Bi-directional Weighted Feature Pyramid, and further improves detection robustness by integrating multiple attention mechanisms into the dynamic detection head. In experiments on a self-built underwater hull surface dataset, the model achieved 87.0% and 62.4% on the mAP@0.5 and mAP@0.5:0.95 metrics, with increases of 2.2% and 3.4% compared to the baseline model, respectively. It provides an accurate and efficient target detection method for underwater hull cleaning tasks, advancing the goal of intelligent and unmanned hull cleaning.