3月30日傍晚，贵州黔东南州“州长杯”榕江赛区闭幕式上，榕江县委书记徐勃走上球场，拿起话筒只说了“闭幕”二字，全程仅6秒，现场先是一愣，随即爆发出热烈掌声与笑声。

4月2日，榕江县体育训练中心主任唐龙向 记者还原当时的场景：当时天已黑、气温低，孩子们踢了一下午球，又累又饿，列队站了很久。徐勃到场后要求简化程序，让孩子们尽早回家用餐休息。

An Adaptive Feature Refinement Network for Maritime Ship Detection in UAV-Based RGB-T ImageryLunxing Wua, Wenyu Xuea, Wenhe Lianga, Yanbo Caoa, Han Wu*b, Kui Liub, Xianghui Zhangb, Kefeng JibaCollege of Electronic Engineering, National University of Defense Technology, Hefei, China; bCollege of Electronic Science and Technology, National University of Defense Technology, Changsha, ChinaABSTRACTSmall ship detection in maritime environments remains challenging due to low illumination, sea-surface reflection, wave clutter, background interference, and the limited saliency of long-range targets. Although RGB and infrared images provide complementary information, existing RGB-T detection methods still suffer from insufficient small-target representation, background redundancy, and limited adaptability to geometric variations. To address these issues, a lightweight improved detection method based on YOLO11-RGBT is proposed for maritime RGB–infrared ship detection. Specifically, an SE attention module is introduced before feature concatenation in the neck to recalibrate channel responses and enhance target-related information during multi-scale fusion, while a DCNv2 module is embedded into deep feature extraction layers to improve the modeling ability for scale variation, shape deformation, and partial occlusion. Experiments are conducted on the offshore ship subset constructed from the RGBT-Tiny dataset. The proposed method achieves a Precision of 0.9707, a Recall of 0.9201, an mAP@0.5 of 0.9707, and an mAP@0.5:0.95 of 0.7361, outperforming RGB-only, infrared-only, and baseline RGBT models. Ablation results further demonstrate that both SE attention and DCNv2 contribute positively, and their combination yields the best overall performance. The proposed method provides an effective and lightweight solution for robust weak-small ship detection in complex maritime multimodal scenarios. Keywords: maritime ship detection, RGB–infrared fusion, YOLO11-RGBT, squeeze-and-excitation attention, DCNv2; multimodal object detection1.INTRODUCTION With the rapid development of maritime transportation, offshore resource exploitation, and unmanned maritime platforms, ship detection in complex ocean environments has become a key enabling technology for intelligent navigation, maritime surveillance, and search-and-rescue systems. However, maritime scenes are often affected by low illumination, sea fog, backlighting, strong reflections, wave clutter, and unstable backgrounds, which significantly reduce target contrast and blur object boundaries. These challenges become even more severe for long-range maritime targets, which usually appear as weak and small objects with limited texture and shape cues, making reliable detection highly difficult.Conventional single-modal ship detection methods still suffer from inherent sensing limitations in maritime scenarios. Visible images preserve rich texture, contour, and structural details, but their performance is highly sensitive to illumination variation and adverse weather. Infrared images, in contrast, are more robust to low-light conditions and can better highlight thermal responses of 

原告现年__69__岁，虽已年满60周岁，但仍具备劳动能力，长期从事务工工作，靠自身劳动获取收入维持生活。2025年___5_月__29__日，被告因交通事故、人身损害、合同违约等行为，导致原告身体受伤，无法正常务工，误工期限自___2025_年___5_月__29__日起至___2026_年___3_月___25_日止，共计误工__300__天。