计算机工程与应用2026，Vol.62Issue(1)：162-171,10.DOI:10.3778/j.issn.1002-8331.2503-0039
基于改进YOLOv11的SAR图像小目标船舶检测
SAR Image Small Target Ship Detection Based on Improved YOLOv11
摘要
Abstract
Complex sea state background,multi-scale characteristics of targets and sensor noise pose great challenges to ship detection based on synthetic aperture radar(SAR)images.Therefore,a small target ship detection model based on improved YOLOv11 SAR images is proposed.A new backbone network is constructed by using depth-separable convolu-tion to reduce the computation and parameter number.SE attention mechanism is used to replace C2PSA attention and enhance feature extraction ability.The residual spatial channel reconstruction convolutional module is designed to replace C3K2 and improve the feature representation ability.Using NWD loss function instead of CIoU loss function makes model inference more focused on small targets.The experimental results show that the mAP@0.5 and recall rate of the improved model on HRSID dataset are 1.5 percentage points and 1.4 percentage points higher than that of YOLOv11 respectively.The parameter number is 1.39×107,and the reasoning time is 0.51 s.Compared with mainstream models,the improved model is superior in both accuracy and speed.In addition,the generalization test of the model on SSDD and RSDD data sets also achieves good results.In summary,the improved model has application potential and popularization value in the small target ship detection task in SAR images.关键词
YOLOv11/合成孔径雷达(SAR)图像/船舶检测/深度可分离卷积/残差空间通道重建卷积/归一化曼哈顿距离(NWD)损失函数Key words
YOLOv11/synthetic aperture radar(SAR)images/ship detection/depth-separable convolution/residual spa-tial channel reconstruction convolution/normalized Wesserstein distance(NWD)loss function分类
信息技术与安全科学引用本文复制引用
艾君鹏,蒋海军,罗亮,郝连东,王仕杰..基于改进YOLOv11的SAR图像小目标船舶检测[J].计算机工程与应用,2026,62(1):162-171,10.基金项目
国家自然科学基金(52101368) （52101368）
广西重大科技专项(桂科AA23062037) （桂科AA23062037）
贵州省交通运输厅科技立项项目(2023-222-033). （2023-222-033）
