正在加载图片...
第17卷第1期 智能系统学报 Vol.17 No.1 2022年1月 CAAI Transactions on Intelligent Systems Jan.2022 D0:10.11992/tis.202107019 网络出版地址:http:/kns.cnki.net/kcms/detail/23.1538.TP.20211217.1715.006html 基于背景建模的VideoSAR动目标阴影检测方法 王鑫',田甜2,田金文2 (1.华中科技大学人工智能与自动化学院,湖北武汉430074;2.华中科技大学多谱信息处理技术国家级重点 实验室,湖北武汉430074) 摘要:针对视频合成孔径雷达(video synthetic aperture radar,.VideoSAR)数据进行地面运动目标检测的问题,本 文提出了一种基于单高斯背景模型的VideoSAR动目标阴影检测方法。该方法使用一个时间维度的滑窗对视 频序列进行处理:首先使用RED20深度神经网络模型抑制VideoSAR图像的斑点噪声,随后使用帧间配准算法 快速配准窗口内的图像序列,然后对序列进行建模和差分得到窗口末帧的二值化前景,最后通过连通区域筛选 和区域生长剔除虚假目标。采用美国Sandia国家实验室公布的VideoSAR视频对本文算法进行了验证,实验表 明,该算法能实现对动目标阴影的准确检测。 关键词:视频合成孔径雷达;动目标检测;卷积神经网络:乘性噪声;去噪;图像配准;单高斯模型:区域生长 中图分类号:TP391 文献标志码:A文章编号:1673-4785(2022)01-0059-10 中文引用格式:王鑫,田甜,田金文.基于背景建模的VideoSAR动目标阴影检测方法J.智能系统学报,2022,17(1):59-68.。 英文引用格式:WANG Xin,TIAN Tian,,TIAN Jinwen.Moving target shadow detection in VideoS..AR based on background model- ingJ CAAI transactions on intelligent systems,2022,17(1):59-68. Moving target shadow detection in VideoSAR based on background modeling WANG Xin',TIAN Tian2,TIAN Jinwen'2 (1.School of Artificial Intelligence and Automation,Huazhong University of Science and Technology,Wuhan 430074,China;2.Na- tional Key Laboratory of Science and Technology on Multi-spectral Information Processing.Huangzhong University of Science and Technology,Wuhan 430074,China) Abstract:Aiming at the problem of ground moving target detection based on Video Synthetic Aperture Radar(Video- SAR)data,a VideoSAR moving target shadow detection method based on single Gaussian background model is pro- posed in this paper,which uses a time-dimensional sliding window to process the video sequence:The RED20 deep neural network model is first used to suppress the speckle noise of VideoSAR image,and then the interframe registra- tion algorithm is applied to quickly register the image sequence of the window.After that,the binary foreground of the last frame of the window is obtained by sequence modeling and background subtraction.Finally,false targets are elimin- ated by connected region screening and region growing.The proposed approach is validated on the VideoSAR video published by Sandia National Laboratory,and experimental results show that the algorithm can accurately detect the shadow of moving targets. Keywords:VideoSAR;moving target detection;convolutional neural network;multiplicative noise;denoising;image registration;single Gaussian model;region growing 针对传统合成孔径雷达(synthetic aperture radar,.SAR)成像帧率较低且最小可检测速度较大 的问题,美国Sandia国家实验室2003年提出了 收稿日期:2021-07-12.网络出版日期:2021-12-21. 基金项目:国家自然科学基金项目(42071339). VideoSAR成像模式,该模式的成像结果类似于 通信作者:田甜.E-mail:tian@hust.edu..cn 视频,能够实现对地面场景的高帧率、高分辨率DOI: 10.11992/tis.202107019 网络出版地址: http://kns.cnki.net/kcms/detail/23.1538.TP.20211217.1715.006.html. 基于背景建模的 VideoSAR 动目标阴影检测方法 王鑫1 ,田甜1,2,田金文1,2 (1. 华中科技大学 人工智能与自动化学院,湖北 武汉 430074; 2. 华中科技大学 多谱信息处理技术国家级重点 实验室,湖北 武汉 430074) 摘 要:针对视频合成孔径雷达 (video synthetic aperture radar,VideoSAR) 数据进行地面运动目标检测的问题,本 文提出了一种基于单高斯背景模型的 VideoSAR 动目标阴影检测方法。该方法使用一个时间维度的滑窗对视 频序列进行处理:首先使用 RED20 深度神经网络模型抑制 VideoSAR 图像的斑点噪声,随后使用帧间配准算法 快速配准窗口内的图像序列,然后对序列进行建模和差分得到窗口末帧的二值化前景,最后通过连通区域筛选 和区域生长剔除虚假目标。采用美国 Sandia 国家实验室公布的 VideoSAR 视频对本文算法进行了验证,实验表 明,该算法能实现对动目标阴影的准确检测。 关键词:视频合成孔径雷达;动目标检测;卷积神经网络;乘性噪声;去噪;图像配准;单高斯模型;区域生长 中图分类号:TP391 文献标志码:A 文章编号:1673−4785(2022)01−0059−10 中文引用格式:王鑫, 田甜, 田金文. 基于背景建模的 VideoSAR 动目标阴影检测方法 [J]. 智能系统学报, 2022, 17(1): 59–68. 英文引用格式:WANG Xin, TIAN Tian, TIAN Jinwen. Moving target shadow detection in VideoSAR based on background model￾ing[J]. CAAI transactions on intelligent systems, 2022, 17(1): 59–68. Moving target shadow detection in VideoSAR based on background modeling WANG Xin1 ,TIAN Tian1,2 ,TIAN Jinwen1,2 (1. School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan 430074, China; 2. Na￾tional Key Laboratory of Science and Technology on Multi-spectral Information Processing, Huangzhong University of Science and Technology, Wuhan 430074, China) Abstract: Aiming at the problem of ground moving target detection based on Video Synthetic Aperture Radar (Video￾SAR) data, a VideoSAR moving target shadow detection method based on single Gaussian background model is pro￾posed in this paper, which uses a time-dimensional sliding window to process the video sequence: The RED20 deep neural network model is first used to suppress the speckle noise of VideoSAR image, and then the interframe registra￾tion algorithm is applied to quickly register the image sequence of the window. After that, the binary foreground of the last frame of the window is obtained by sequence modeling and background subtraction. Finally, false targets are elimin￾ated by connected region screening and region growing. The proposed approach is validated on the VideoSAR video published by Sandia National Laboratory, and experimental results show that the algorithm can accurately detect the shadow of moving targets. Keywords: VideoSAR; moving target detection; convolutional neural network; multiplicative noise; denoising; image registration; single Gaussian model; region growing 针对传统合成孔径雷达 (synthetic aperture radar, SAR) 成像帧率较低且最小可检测速度较大 的问题,美国 Sandia 国家实验室 2003 年提出了 VideoSAR[1] 成像模式,该模式的成像结果类似于 视频,能够实现对地面场景的高帧率、高分辨率 收稿日期:2021−07−12. 网络出版日期:2021−12−21. 基金项目:国家自然科学基金项目 (42071339). 通信作者:田甜. E-mail: ttian@hust.edu.cn. 第 17 卷第 1 期 智 能 系 统 学 报 Vol.17 No.1 2022 年 1 月 CAAI Transactions on Intelligent Systems Jan. 2022
向下翻页>>
©2008-现在 cucdc.com 高等教育资讯网 版权所有