船载HFSWR船只目标多维样本库构建方案设计  

Design of a multidimensional vessel target sample database construction scheme for shipborne HFSWR

作　　者：纪永刚 任继红 李发瑞 李桃利 王佳伟 JI Yonggang;REN Jihong;LI Farui;LI Taoli;WANG Jiawei(College of Oceanography and Space Informatics,China University of Petroleum(East China),Qingdao 266580,China;Technology Innovation Center for Maritime Silk Road Marine Resources and Environment Networked Observation,Ministry of Natural Resources,Qingdao 266580,China)

机构地区：[1]中国石油大学(华东)海洋与空间信息学院,山东青岛266580 [2]自然资源部海上丝路海洋资源环境组网观测技术创新中心,山东青岛266580

出　　处：《实验技术与管理》2025年第1期19-27,共9页Experimental Technology and Management

基　　金：国家自然科学基金(62271507,62031015);山东省自然科学基金资助项目(ZR2022MF235)。

摘　　要：深度学习方法在船载高频地波雷达(high-frequency surface wave radar,HFSWR)船只目标检测中得到应用,然而其成功的关键在于构建精准全面的目标样本库。当前基于距离-多普勒(range-Doppler,RD)谱检测结果的单一维度样本库不完备,限制了模型的学习效果。为解决该问题,文章设计了一种船载HFSWR船只目标多维样本库构建方案,包含3个维度:①基于RD谱的检测结果结合自动识别系统(automatic identification system,AIS)验证RD维度;②基于多帧航迹段识别RT(range-time)/DT(Doppler-time)维度;③基于时频TF(time-frequency)分析检测TF维度。通过对3个维度的匹配和重合目标筛除以构建样本集。最后,利用实测数据构建的目标样本库对基于U-Net的目标检测网络进行训练和测试,结果表明该样本库可满足基于深度学习的目标检测模型训练需求。[Objective]Shipborne high-frequency surface wave radar(HFSWR)offers exceptional capabilities for over-the-horizon tracking and monitoring of vessel targets.The advantages of shipborne HFSWR include a scalable detection area and independence from fixed locations.However,most existing methods for constructing vessel target sample databases rely mainly on range-Doppler(RD)spectra.Therefore,these databases are often incomplete and fail to adequately represent weak signal targets,particularly in complex environments.This deficiency hinders the training of deep learning models,further reducing their effectiveness and generalizability in the real world.[Methods]To develop a comprehensive and accurate dataset,a vessel target sample database construction scheme is proposed.This scheme integrates multidimensional information,encompassing the RD,range-time/Doppler-time(RT/DT),and time-frequency(TF)dimensions.In the RD dimension,target samples are generated by matching automatic identification system(AIS)data with RD spectral detection results.Targets are added to the sample database if the matching is successful.This method is particularly effective for high signal-to-noise ratio(SNR)targets because AIS data enhance the target localization accuracy.For targets in the RD dimension that lack corresponding AIS data,the dynamic programming-track before detect(DP-TBD)algorithm is applied within the RT/DT dimension,which analyzes track segments across multiple frames of RD spectra to validate the authenticity of the targets.If continuous track segments are identified across multiple frames,the target is added to the sample database.In the TF dimension,a time-frequency analysis method is employed to detect weak targets that may not be visible in the RD spectra.By extracting time-frequency ridge information,data related to these weak targets are incorporated into the sample database.Finally,by integrating information from these dimensions,a diverse and comprehensive vessel target sample dataset is created,including strong signal ta

关 键 词：船载HFSWR 船只目标检测 多维度信息融合 目标样本库构建 深度学习 

分 类 号：TN958.93[电子电信—信号与信息处理]