基于深度特征的质量感知旋转舰船模板匹配算法

doi:10.19678/j.issn.1000-3428.0066579

摘要/Abstract

摘要：

天基高价值目标连续跟踪在态势预警、军事决策等方面有着广泛应用。目前基于遥感视频卫星的跟踪都是单星目标跟踪，由于低轨卫星过顶时间有限，因此难以进行连续跟踪。为了实现跨星接力跟踪，接力卫星在经过非连续时间的目标环境和观察视角变化后需要唯一地匹配出高价值目标，为此，采用卷积神经网络VGG19作为主干网络，提出一种基于深度特征的质量感知旋转舰船模板匹配算法。设计双级特征融合模块，通过融合不同深度的特征信息解决因舰船目标尺寸差异较大导致的匹配不准确问题；针对因环境变化导致的定位不准确问题，引入质量感知模板匹配模块；使用细粒度舰船角度定位模块，利用贝叶斯公式确定舰船方向信息，为跨星连续跟踪提供准确的模板。实验结果表明，该算法提高了舰船目标匹配准确率，相对于QATM、DDIS、SIFT方法，AUC分别提升了9.5、16.0和17.5个百分点，AP75分别提升了21.1、30.5和6.9个百分点，所提算法可以有效提高舰船模板匹配的精度，能够为实现卫星星座跨星连续跟踪提供技术支持。

关键词: 深度特征, 卷积神经网络, 模板匹配, 特征融合, 遥感图像

Abstract:

Space-based high-value-target continuous tracking has a wide range of applications in situation warning and military decision-making. Currently, single-satellite target tracking is used for the remote-sensing video satellite; however, it is difficult to track continuously because of the limited overhead time of low-earth-orbit satellite. Therefore, it is necessary for relay satellites to uniquely match high-value targets after noncontinuous time changes occur in the target environment and viewing angle to achieve cross-satellite relay tracking. This paper proposes a quality-aware rotating-ship template matching algorithm based on deep features using the Convolutional Neural Network(CNN)VGG19 backbone network. A dual-level feature fusion module is designed to solve the problem of inaccurate matching caused by significant differences in the ship target size by fusing feature information from different depths. A quality-aware template matching module is introduced to solve the problem of inaccurate positioning caused by environmental changes. Finally, the fine-grained ship angle-positioning module is used to determine the ship orientation information using the Bayesian formula, which provides an accurate template for cross-satellite continuous tracking. The experimental results show that the proposed method improves the accuracy of ship target matching. Compared with the AUC values obtained using the QATM, DDIS, and SIFT methods, the AUC of the proposed method increases by 9.5, 16.0, and 17.5 percentage points, respectively, and the AP75 increases by 21.1, 30.5, and 6.9 percentage points, respectively. The proposed method effectively improves the accuracy of ship template matching and is valuable for achieving the cross-satellite continuous tracking of satellite constellations.

Key words: deep features, Convolutional Neural Network(CNN), template matching, feature fusion, remote sensing images

王泽瑞, 陈实. 基于深度特征的质量感知旋转舰船模板匹配算法[J]. 计算机工程, 2023, 49(12): 161-168.

Zerui WANG, Shi CHEN. Quality-Aware Rotating-Ship Template Matching Algorithm Based on Deep Features[J]. Computer Engineering, 2023, 49(12): 161-168.

http://www.ecice06.com/CN/Y2023/V49/I12/161

图/表 12

图1 卫星星座接力跟踪处理流程

Fig.1 Relay tracking process of satellite constellation

图2 Rotated QATM方法流程

Fig.2 Procedure of the Rotated QATM method

图3 双级特征融合模块结构

Fig.3 Dual-level feature fusion module structure

图4 目标定位方法

Fig.4 Target positioning method

图5 旋转角度定位方法

Fig.5 Rotation angle positioning method

图6 4种算法的准确率曲线

Fig.6 Accuracy curves of four algorithms

图7 4种算法的PR曲线对比

Fig.7 Comparison of PR curves of four algorithms

图8 不同特征层匹配的ROC曲线对比

Fig.8 Comparison of ROC curves of different feature layer matching

图9 3组存在典型干扰的匹配对的实验结果

Fig.9 Experimental results of three paired pairs with typical interference

参考文献 25

1	WOOSTER M J, ROBERTS G J, GIGLIO L, et al. Satellite remote sensing of active fires: history and current status, applications and future requirements. Remote Sensing of Environment, 2021, 267, 112694. doi: 10.1016/j.rse.2021.112694
2	LI D, WANG M, JIANG J. China's high-resolution optical remote sensing satellites and their mapping applications. Geo-Spatial Information Science, 2021, 24(1): 85- 94. doi: 10.1080/10095020.2020.1838957
3	CUI Y, HOU B, WU Q, et al. Remote sensing object tracking with deep reinforcement learning under occlusion. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60(1): 1- 13.
4	KOTHAWADE S, ROY D, FENZI M, et al. Object-level targeted selection via deep template matching[EB/OL]. [2022-11-05]. https://arxiv.org/abs/2207.01778.
5	朱辉, 秦品乐. 基于多尺度特征结构的U-Net肺结节检测算法. 计算机工程, 2019, 45(4): 254- 261. doi: 10.19678/j.issn.1000-3428.0051769
	ZHU H, QIN P L. U-Net pulmonary nodule detection algorithm based on multi-scale feature structure. Computer Engineering, 2019, 45(4): 254- 261. doi: 10.19678/j.issn.1000-3428.0051769
6	REN Q, ZHENG Y, SUN P, et al. A robust and accurate end-to-end template matching method based on the siamese network. Geoscience and Remote Sensing Letters, 2021, 19, 1- 5.
7	ORON S, DEKEL T, XUE T F, et al. Best-buddies similarity-robust template matching using mutual nearest neighbors. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(8): 1799- 1813. doi: 10.1109/TPAMI.2017.2737424
8	王自全, 张永生, 于英, 等. 深度学习背景下视觉显著性物体检测综述. 中国图象图形学报, 2022, 27(7): 2112- 2128. URL
	WANG Z Q, ZHANG Y S, YU Y, et al. Review of deep learning based salient object detection. Journal of Image and Graphics, 2022, 27(7): 2112- 2128. URL
9	SARDAR M S, ALAEIYAN M, FARAHANI M R, et al. Resistance distance in some classes of rooted product graphs obtained by Laplacian generalized inverse method. Journal of Information and Optimization Sciences, 2021, 42(7): 1447- 1467. doi: 10.1080/02522667.2021.1899210
10	常洪彬, 李文举, 李文辉. 基于注意力机制的航空图像旋转框目标检测. 吉林大学学报(理学版), 2022, 60(6): 1363- 1369. URL
	CHANG H B, LI W J, LI W H. Rotated object detection in aerial images based on attention mechanism. Journal of Jilin University(Science Edition), 2022, 60(6): 1363- 1369. URL
11	PEREK S, HAZAN A, BARKAN E, et al. Siamese network for dual-view mammography mass matching[EB/OL]. [2022-11-05]. https://link.springer.com/content/pdf/10.1007/978-3-030-00946-5_6.pdf.
12	JATMOKO C, SINAGA D, SUGIARTO E, et al. Pattern recognition on vehicle number plates using a fast match algorithm. Journal of Applied Intelligent System, 2021, 6(2): 103- 110. doi: 10.33633/jais.v6i2.4625
13	ZHANG C, AKASHI T. Fast affine template matching over Galois field[EB/OL]. [2022-11-05]. https://www.researchgate.net/publication/285249166_Fast_Affine_Template_Matching_over_Galois_Field.
14	CHENG J, WU Y, ABD-ALMAGEED W, et al. QATM: quality-aware template matching for deep learning[EB/OL]. [2022-11-05]. https://arxiv.org/abs/1903.07254.pdf.
15	KAT R, JEVNISEK R, AVIDAN S. Matching pixels using co-occurrence statistics[C]//Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2018: 1751-1759.
16	TALMI I, MECHREZ R, ZELNIK-MANOR L. Template matching with deformable diversity similarity[EB/OL]. [2022-11-05]. https://arxiv.org/abs/1612.02190.pdf.
17	ORON S, DEKEL T, XUE T F, et al. Best-buddies similarity-robust template matching using mutual nearest neighbors. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(8): 1799- 1813. doi: 10.1109/TPAMI.2017.2737424
18	李熙莹, 周智豪, 邱铭凯. 基于部件融合特征的车辆重识别算法. 计算机工程, 2019, 45(6): 12- 20. URL
	LI X Y, ZHOU Z H, QIU M K. Vehicle re-identification algorithm based on component fusion feature. Computer Engineering, 2019, 45(6): 12- 20. URL
19	ZHANG L, ZHANG P. Target tracking algorithm for template transformation based on template matching[C]//Proceedings of 2020 IEEE International Conference on Mechatronics and Automation. Washington D. C., USA: IEEE Press, 2020: 1659-1663.
20	REN Q, ZHENG Y, SUN P, et al. A robust and accurate end-to-end template matching method based on the siamese network. Geoscience and Remote Sensing Letters, 2021, 19, 18- 24.
21	GEIRHOS R, RUBISCH P, MICHAELIS C, et al. ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness[EB/OL]. [2022-11-05]. https://arxiv.org/abs/1811.12231.pdf.
22	ZHAO Y, SHI Y, WANG Z. The improved YOLOv5 algorithm and its application in small target detection[C]//Proceedings of International Conference on Intelligent Robotics and Applications. Berlin, Germany: Springer, 2022: 679-688.
23	YE Y, BRUZZONE L, SHAN J, et al. Fast and robust matching for multimodal remote sensing image registration[EB/OL]. [2022-11-05]. https://arxiv.org/abs/1808.06194.pdf.
24	QING Y, LIU W, FENG L, et al. Improved YOLO network for free-angle remote sensing target detection. Remote Sensing, 2021, 13(11): 2171- 2183.
25	CARVALHO T, DE REZENDE E R S, ALVES M T P, et al. Exposing computer generated images by eye's region classification via transfer learning of VGG19 CNN[C]//Proceedings of 2017 IEEE International Conference on Machine Learning and Applications. Washington D. C., USA: IEEE Press, 2017: 866-870.

[1]	范文卓, 吴涛, 许俊平, 李庆庆, 张建林, 李美惠, 魏宇星. 基于多分辨率特征融合的任意尺度图像超分辨率重建[J]. 计算机工程, 2023, 49(9): 217-225.
[2]	朱孟栩, 张文豪, 李国洪, 顾行发, 余涛, 郑逢杰, 张丽丽, 吴俣, 邴芳飞, 唐健雄. 基于卷积神经网络的高分六号卫星多光谱图像压缩[J]. 计算机工程, 2023, 49(9): 287-294.
[3]	李现国, 李滨. 基于Transformer和多尺度CNN的图像去模糊[J]. 计算机工程, 2023, 49(9): 226-233, 245.
[4]	李嘉新, 侯进, 盛博莹, 周宇航. 基于改进YOLOv5的遥感小目标检测网络[J]. 计算机工程, 2023, 49(9): 256-264.
[5]	杜逸潇, 王红军, 李修和. 基于频谱地图的辐射源指纹定位方法研究[J]. 计算机工程, 2023, 49(9): 183-190, 198.
[6]	胡水. 基于深度强化学习的智能兵棋推演决策方法[J]. 计算机工程, 2023, 49(9): 303-312.
[7]	韩璐, 霍纬纲, 张永会, 刘涛. 基于多尺度特征融合与双注意力机制的多元时间序列预测[J]. 计算机工程, 2023, 49(9): 99-108.
[8]	李哲铭, 王晋东, 侯建中, 李伟, 张世华, 张恒巍. 基于显著区域优化的对抗样本攻击方法[J]. 计算机工程, 2023, 49(9): 246-255, 264.
[9]	宋志娜, 李莎, 杨建明, 徐川. 基于特征与区域定位增强的遥感舰船目标检测[J]. 计算机工程, 2023, 49(8): 257-264.
[10]	张欣怡, 张飞, 郝斌, 高鹭, 任晓颖. 基于改进YOLOv5的口罩佩戴检测算法[J]. 计算机工程, 2023, 49(8): 265-274.
[11]	杨长沛, 廖列法. 基于门控空洞卷积特征融合的中文命名实体识别[J]. 计算机工程, 2023, 49(8): 85-95.
[12]	杨祖赫, 黎智辉, 唐云祁, 晏于文, 宋华青. 结合语义与图像信息的行人属性识别算法[J]. 计算机工程, 2023, 49(8): 215-222, 231.
[13]	陈露萌, 曹彦彦, 黄民, 谢鑫钢. 基于改进YOLOv5的火焰检测方法[J]. 计算机工程, 2023, 49(8): 291-301, 309.
[14]	曹坪, 杨怀志, 薄一军, 尤嘉, 张淳杰, 李丹勇. 面向低质量裂缝图像的多知识蒸馏分类[J]. 计算机工程, 2023, 49(7): 204-213.
[15]	白明昌. 基于折叠路径聚合的属性网络节点嵌入方法[J]. 计算机工程, 2023, 49(7): 76-84.

选择文件类型/文献管理软件名称

选择包含的内容