改进FCOS的SAR图像舰船检测算法  

SAR image ship detection algorithm based on improved FCOS

作　　者：桑林[1] Sang Lin(College of Innovation&Entrepreneurship,Heilongjiang University of Science&Technology,Harbin 150022,China)

机构地区：[1]黑龙江科技大学创新创业学院,哈尔滨150022

出　　处：《黑龙江科技大学学报》2024年第4期637-641,共5页Journal of Heilongjiang University of Science And Technology

摘　　要：针对SAR图像中舰船检测的目标尺度变化大及背景复杂等影响因素,提出一种基于FCOS的一阶段舰船目标检测算法。采用基于拆分注意力和分组卷积的ResNeSt网络作为主干网络进行提取特征,同时在特征金字塔基础上增加聚合路径和注意力机制,提升特征融合能力,实现对网络结构的优化。结果表明,改进方法相对于基线网络平均精度提升了2.15%,精准率提升了2.4%,召回率提升了3.59%。该研究在处理SAR图像中舰船检测任务时具有较好的性能。This paper proposes a one-stage ship target detection algorithm based on FCOS to address the ship detection influence factors of the significant changes of target scale and the complex background in SAR images.The study consists of using ResNeSt network based on split attention and group convolution as the backbone network to extract features;meanwhile,adding aggregation paths and attention mechanisms on the basis of the feature pyramid to improve feature fusion ability and optimize the network structure.The results show that the improved method improves the average accuracy of the baseline network by 2.15%,accuracy by 2.4%,and recall by 3.59%.This proposed method has good performance in processing ship detection tasks in SAR images.

关 键 词：目标识别 SAR图像 舰船检测 FPN 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]