厦门大学学报（自然科学版）2025，Vol.64Issue(6)：949-957,9.DOI:10.6043/j.issn.0438-0479.202506008
基于改进YOLOv8的SAR图像智能识别方法
SAR image intelligent recognition method based on improved YOLOv8
摘要
Abstract
[Objective]Due to the fact that synthetic aperture radar(SAR)images are sensitive to shooting angles and target postures,it is difficult for the model to distinguish targets in SAR images.In this article,we propose an SAR image vehicle target intelligent recognition model DCF-SAR based on YOLOv8.[Method]First,in the model,a deformable convolutional network(DCN)module is introduced to enhance its ability to recognize deformed targets.Second,by introducing the content-aware reassembly of features(CARAFE)upsampling module,the efficiency of the model in the feature fusion stage can be improved.Finally,for the purpose of further enhancing the training efficiency and feature extraction ability of the model,the Focal loss function is used to improve the cross entropy classification loss function of the original model,so that the classification accuracy of the model and its attention to the target area can be improved.[Results]Experimental results on the MSTAR dataset show that the DCF-SAR model achieves a recognition accuracy of 98.79％on the extend operating condition(EOC)dataset with large depression angle variations,which is 0.12 percentage points higher than the original model.On the standard operating condition(SOC)dataset with only azimuth angle variations and at 1/12 the scale,it achieves a recognition accuracy of 89.13％,significantly improving the original model by 10.98 percentage points.[Conclusion]The DCF-SAR model is not only suitable for scenarios with large depression angle variations but also demonstrates outstanding performance on small-scale datasets with limited azimuth angle changes.关键词
合成孔径雷达/可变形卷积网络/内容感知特征重组/Focal损失函数Key words
synthetic aperture radar/deformable convolutional network/content-aware reassembly of features/Focal loss function分类
信息技术与安全科学引用本文复制引用
闫晨宇,耿亮,杜伟伟..基于改进YOLOv8的SAR图像智能识别方法[J].厦门大学学报（自然科学版）,2025,64(6):949-957,9.基金项目
陆军装备部"十四五"预先研究基金资助项目(804) （804）
