文献类型：期刊文献

英文题名：YOLOv11n-SSS: A dynamic dual-branch attention detection model for sea urchins in marine ranching

作者：Qiao, Lijie[1,2];Wang, Ji[2,3]

机构：[1]Guangdong Ocean Univ, Sch Math & Comp Sci, Zhanjiang 524088, Guangdong, Peoples R China;[2]Guangdong Ocean Univ, Guangdong Prov Engn Technol Res Ctr Smart Marine S, Zhanjiang 524088, Guangdong, Peoples R China;[3]Guangdong Ocean Univ, Sch Elect & Informat Engn, Zhanjiang 524088, Guangdong, Peoples R China

年份：2026

卷号：363

外文期刊名：OCEAN ENGINEERING

收录：SCI-EXPANDED(收录号：WOS:001846429100001)、、EI(收录号：20263221269747)、Scopus(收录号：2-s2.0-105046714219)、WOS

基金：This research is funded by the New Generation Information Tech-nology Special Project in Key Fields of Guangdong Province's General Universities (Grant No. 2020ZDZX3008) .

语种：英文

外文关键词：Underwater sea urchin detection; YOLOv11; Spine-aware background filtering; Space-to-depth convolution; Dynamic weighted loss; Marine ranching

外文摘要：Accurate detection of underwater sea urchins is important for intelligent marine ranching, but it remains challenging because radial spine textures are easily confused with complex backgrounds, fine-grained features of juvenile individuals are often degraded during down-sampling, and dense aggregation can cause occlusions and missed detections. Existing general-purpose detectors rarely model the morphological characteristics of sea urchins or the spatial relationships among densely overlapping individuals. To address these challenges, a lightweight detection model, YOLOv11n-SSS, is developed for underwater sea urchin detection. First, a Spine-Aware Background-Filtering module is designed, comprising an edge-enhancement branch and a global backgroundfiltering branch, to enhance spine-related responses while reducing interference from visually similar underwater backgrounds. Second, Space-to-Depth Convolution is selectively integrated into shallow down-sampling stages, transferring spatial information into the channel dimension and thereby mitigating the loss of finegrained features in juvenile sea urchins. Third, a Scale-Aware Dynamic Weighted Loss is proposed to assign greater importance to small targets and introduce additional localization penalties for densely overlapping instances based on the spatial overlap relationships. Ablation experiments indicate that these three components yield complementary improvements. Compared with the YOLOv11n baseline, YOLOv11n-SSS improves precision, recall, mAP@50, and mAP@50:95 by 1.7, 1.1, 1.4, and 1.2 percentage points, respectively, while adding only 0.29 M parameters and 2.4 GFLOPs. These results indicate that YOLOv11n-SSS achieves a favorable balance between detection accuracy and computational efficiency, providing a practical approach for real-time sea urchin monitoring in intelligent marine ranching. Our code is available at https://github.com/Qiao12-pixel/YOL Ov11n-SSS.