哈尔滨商业大学学报（自然科学版）2025，Vol.41Issue(5)：515-522,8.
基于改进YOLOv8的SAR舰船检测算法
Improved algorithm for detecting ship target in SAR images based on YOLOv8 model
摘要
Abstract
Aiming at the problems of low detection accuracy,missed detection,large model size and difficulty in deployment,and high computational cost in ship target detection tasks of SAR images,this paper proposed an improved ship target detection algorithm,YOLOv8+ADown+EMA.The ADown module was used to improve YOLOv8,reducing the spatial dimension of the feature map,so that it can better capture image features and achieve the goal of reducing the model's computational cost.An EMA module was introduced into the neck,and attention adjustment was performed in the feature fusion stage to achieve precise detection.On the SSDD dataset,the proposed algorithm achieved a 3.7%,5.6%,and6.4%improvement in prediction rate,recall rate,and mAP@0.5,respectively,compared with the original algorithm,significantly enhancing the model's ability to detect ship targets in SAR images.关键词
SAR图像/舰船目标检测/YOLOv8/EMAKey words
SAR image/ship target detection/YOLOv8/EMA分类
信息技术与安全科学引用本文复制引用
敖雪凌霜,拉希姆,沈众卫,舒文一..基于改进YOLOv8的SAR舰船检测算法[J].哈尔滨商业大学学报（自然科学版）,2025,41(5):515-522,8.基金项目
国家重点研发计划(2022YFE0136800) （2022YFE0136800）
中国船舶工业综合技术经济研究院海洋防务技术创新基金(JJ-2022-719-03) （JJ-2022-719-03）
微系统技术国防科技重点实验室开放课题(6142804230106) （6142804230106）
自然资源部海洋环境探测技术与应用重点实验室开放基金(MESTA-2022A006) （MESTA-2022A006）
黑龙江省2023年新一轮"双一流"学科协同创新成果项目(LJGXCG2023-066) （LJGXCG2023-066）
智能跨水空介质通信机产业化及示范(CXRC20231113756) （CXRC20231113756）
哈尔滨工程大学南海研究院南海高水平科技创新引导专项:基于低轨卫星的深远海跨域信息传输与组网技术 （）
