Ship Target Detection Model Based on Improved YOLOv11

Ship detection in UAV aerial imagery is complicated by small, densely clustered targets, large scale variation, and cluttered ocean backgrounds. This paper presents MDW-YOLO, a YOLOv11-based detector addressing these challenges through three modifications. The MBFC module replaces standard Conv blocks in the neck, using parallel heterogeneous branches to jointly enrich spatial texture and channel semantics for small targets. DA- Attention combines coordinate channel and spatial attention in parallel, improving small-target recognition under complex backgrounds by capturing long-range dependencies without sacrificing spatial precision. Finally, CIoU loss is replaced by WIoU, whose dynamic focusing mechanism down-weights degenerate anchors and sharpens sensitivity to blurred or partially occluded boundaries. On SeaDronesSee, MDW-YOLO gains 3.7 pp on mAP50 and 6.3 pp on mAP50:95 over YOLOv11n, with only 0.2 M additional parameters. Comparisons against RTDETR, YOLOv8s, and other detectors confirm the approach's effectiveness.