摘要
由于合成孔径雷达(Synthetic Aperture Radar)图像中环境复杂、舰船目标特征多样化,传统检测方法通常计算量大,难以在低算力平台实现实时检测。为此,本文提出了一种基于YOLOv11n的轻量化神经网络模型。首先,设计了新的下采样模块(SCG-Down),以提高小目标细节和大尺度语义的特征表达能力;其次,结合新型特征聚合网络(GFPN),增强浅层与深层特征的交互,提升多尺度目标的检测性能。实验在HRSID数据集上进行,结果显示,所提方法的mAP@0.5达到90%,相较于基准模型提升3%,在精度和效率上优于现有方法。
Due to the complex environment and diversity of ship target features in Synthetic Aperture Radar(SAR)images,traditional detection methods typically involve high computational costs,making real-time detection on low-computing-power platforms challenging.To address this ssue,this paper proposes a lightweight neural network model based on YOLOvlln.Firstly,a novel downsampling module(SCG-Down)is designed to enhance the feature representation of small target details and large-scale semantics.Secondly,a new feature aggregation network(GFPN)is introduced to improve the interaction between shallow and deep features,thereby enhancing the detection performance for multi-scale targets.Experiments conducted on the HRSID dataset demonstrate that the proposed method achieves a mAP@0.5 of 90%,representing a 3%improvement over the baseline model.The proposed method outperforms existing approaches in both accuracy and efficiency.
作者
赵梓翔
王樱洁
张华春
Zhao Zixiang;Wang Yingjie;Zhang Huachun(Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing,China;School of Electronic,Electrical and Communication Engineering,University of Chinese Academy of Sciences,Beijing,China)
出处
《科学技术创新》
2025年第20期208-212,共5页
Scientific and Technological Innovation
基金
国家自然科学基金(青年基金)项目(61901444)资助。
