摘要
针对传统舰船目标变化检测方法易受背景干扰、依赖人工特征设计等问题,提出一种基于深度学习的港口舰船目标变化检测方法。首先,对多时相港口影像进行几何配准与辐射校正等预处理,结合舰船停靠规律构建舰船变化兴趣区并提取对应影像;其次,利用FGSD2021数据集训练DDMNet深度学习模型,实现兴趣区内舰船的精准检测;最后,通过计算多时相舰船检测结果的交并比判定目标变化状态。利用两组数据进行实验,变化检测正确率分别达到100%和88.89%,相较于直接应用深度学习目标检测的方法,本方法有效降低了复杂背景的干扰,显著提升了舰船变化检测的准确性。
To address the issues of traditional ship target change detection methods,such as susceptibility to background interference and reliance on manually designed features,this paper proposes an approach based on deep learning for ship change detection in port areas.First,multi-temporal port images undergo preprocessing,including geometric registration and radiometric correction,followed by the construction of ship change regions of interest(ROIs)based on ship docking patterns and the extraction of corresponding image patches.Next,the DDMNet deep learning model is trained using the FGSD2021 dataset to achieve accurate ship detection within the ROIs.Finally,the Intersection over Union(IoU)of ship detection results across multi-temporal images is calculated to determine changes in ship targets.Experimental results on two datasets demonstrate change detection accuracies of 100%and 88.89%,respectively.Compared to direct object detection methods based on deep learning,the proposed approach effectively reduces interference from complex backgrounds and significantly improves the accuracy of ship change detection.
作者
刘卓然
徐俊峰
彭进先
余东行
LIU Zhuoran;XU Junfeng;PENG Jinxian;YU Donghang(School of Geomatics and Urban Spatial Informatics,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;School of Non-Commissioned Officer,Space Engineering University,Beijing 102249,China;63611 tro0ps,Korla 841000,China;Naval Research Institute,Beijing 100070,China)
出处
《海洋测绘》
北大核心
2025年第3期56-60,共5页
Hydrographic Surveying and Charting
基金
国家自然科学基金(62301606)。
关键词
目标变化检测
舰船检测
深度学习
变化兴趣区
交并比
target change detection
ship detection
deep learning
change region of interest
intersection over union
