quotation:[Copy]
[Copy]
【Print page】 【Download 【PDF Full text】 【View/Add Comment】【Download reader】【 Close 】

←Previous page|Page Next →

Back Issue    Advanced search

This Paper:Browse 219   Download 224 本文二维码信息
码上扫一扫!
基于聚焦扩散机制的SAR图像舰船旋转框目标检测
黄洪琼,高杰
0
(上海海事大学 信息工程学院,上海 201306)
摘要:
针对合成孔径雷达(Synthetic Aperture Radar,SAR)图像舰船目标容易被背景和噪声淹没、舰船目标尺度变化大、目标检测定位不准确等问题,提出了一种基于聚焦扩散机制的SAR图像舰船旋转框目标检测方法:FCFO_NET。首先采用非局部平均去噪技术(Fastnlmd)对SAR图像进行预处理,以消除噪声并改善图像质量。其次,优化YOLOv8的主干网络,采用ConvNeXtV2架构的纯卷积神经网络,有效应对复杂背景下检测效果不佳的问题。接着,构建了聚焦扩散的金字塔网络结构,高效捕获并整合跨多个尺度的丰富信息。最后,引入定向边界框(Oriented Bounding Box,OBB)旋转框目标检测头更准确地定位舰船目标。在SSDD+数据集上的实验结果表明,所提方法在平均精度(mAP)上达到98.9%的准确率,F1分数达到96.7%,分别比基线模型提高2.3和3.1个百分点。
关键词:  合成孔径雷达(SAR)  舰船目标检测  聚焦扩散金字塔网络(FDPN)  定向边界框  非局部平均去噪
DOI:10.20079/j.issn.1001-893x.250119002
基金项目:国家自然科学基金面上项目(62271303)
Ship Rotated Bounding Box Target Detection in SAR Images Based on Focusing Diffusion Mechanism
HUANG Hongqiong,GAO Jie
(School of Information Engineering,Shanghai Maritime University,Shanghai 201306,China)
Abstract:
To address the challenges of synthetic aperture radar(SAR) image ship target detection,such as targets being easily submerged by background noise,large-scale changes of ship targets,and inaccurate localization,the authors propose a method for ship rotating frame target detection in SAR images based on focusing diffusion mechanism named FCFO_NET.Firstly,Fast Non-local Means Denoising(Fastnlmd) is used to pre-process SAR images to eliminate noise and improve image quality.Secondly,the backbone network of YOLOv8 is optimized,and a pure convolutional neural network with ConvNeXtV2 architecture is adopted to effectively deal with the problem of poor detection in complex backgrounds.Next,a Focusing Diffusion Pyramidal Network(FDPN) structure is constructed to efficiently capture and integrate rich information across multiple scales.Finally,the oriented bounding box(OBB) rotating frame target detection head is introduced to locate ship targets more accurately.The experimental results on the SSDD+ datasets show that the proposed method achieves a mean average precision(mAP) of 98.9% and an F1 score of 96.7%,which are 2.3 and 3.1 percentage points higher than the baseline model,respectively.
Key words:  synthetic aperture radar(SAR)  ship target detection  focusing diffusion pyramid network(FDPN)  oriented bounding box  fast non-local means denoising