Detecting small objects in open-water UAV imagery is challenging due to low contrast, scale variation, and tight on-board latency constraints. We present RT-DETR-MO, where “MO” stands for Maritime Open-water, a lightweight transformer-based detector tailored for maritime scenarios. The design introduces three targeted components: a Dynamic Inception-style Mixed Convolution block (DiMConv) for adaptive multi-scale representation, a Locally-enhanced Token Statistics Self-Attention (LTSSA) that injects neighborhood priors into linear-time attention to emphasize small or clustered targets, and a lightweight Modulation Fusion Module (MFM) for branch-aware feature integration. On the SeaDronesSee benchmark, RT-DETR-MO achieves 83.9% mAP@50 and 49.9% mAP@50:95, surpassing the RT-DETR baseline by 2.4 and 2.0 points, respectively. It also cuts parameters by 35.7% and boosts inference speed by 40.7%. These results demonstrate a more favorable accuracy–efficiency–size trade-off for real-time maritime UAV detection.