基于深度学习的舰船关键部位检测算法

doi:10.12305/j.issn.1001-506X.2026.02.04

摘要/Abstract

摘要：

针对当前缺乏舰船关键部位的检测算法、对应数据集，检测算法精度速度无法平衡及网络对舰船位置尺度变换鲁棒性不强等技术难题，构建一种三维特征增强和不同尺度特征结合的无锚框的舰船关键部位选择方法，基于相似度特征模块的深层聚合分割算法，实现对舰船关键部位的精准高效检测。首先，通过引入感受野模块实现网络多尺度特征融合，提升检测精度。然后，通过并入基于相似度的注意力模块提升对有用目标信息的关注度；通过使用可变形卷积实现对不同层的特征信息进行聚合，有效提升网络的泛化能力和表达能力。最后，在不使用锚框的前提下，通过目标中心点预测，再回归得到中心点偏移、目标角度、尺度信息，提升检测速度。分别在自建数据集及Pascal视觉对象类别（Visual Object Classes，VOC）数据集上进行对比实验，充分证明了所提网络对舰船关键部位检测的精准性及时效性，能够为反舰装备实现外科手术式打击提供可行技术途径及理论支撑。

关键词: 舰船目标, 关键部位, 相似度注意力机制, 特征融合, 精准选择

Abstract:

Aiming at the problems that critical parts detection algorithm and corresponding data set are lacking, accuracy and speed of detection algorithm can not be balanced, and the network is not robust to ship position scaling, a multi-scale feature fusion and three dimensional feature enhancement method of ship critical parts detection without anchor frame: similarity-based attention module-deep layer aggregation segmentation algorithm is proposed to achieve precise and efficient detection of critical parts of ships. This method can achieve accurate and efficient detection of critical parts of ships. Firstly, the receptive fields block is introduced to realize multi-scale feature fusion and improve the detection accuracy. Then, by incorporating similarity-based attention module, attention to useful target information is improved in the network. by using deformable convolution to aggregate the feature information of different layers, the generalization ability and expression ability of the network are effectively improved. Finally, on the premise of not using anchor frame and improving the detection speed, the network get the center point offset, target angle, scale information through the target center point prediction, and then regression. Comparative experiments are carried out on the self-built data set and PASCAL VOC data set respectively, which fully proved the accuracy and timeliness of the proposed network detection of ship critical parts, and at the proposed network could provide feasible technical approaches and theoretical support for anti-ship equipment to achieve surgical strike.

Key words: ship target, critical part, similarity-based attention module, feature fusion, precise choice

中图分类号:

E 919

王瑶, 胥辉旗, 曹司磊, 王磊. 基于深度学习的舰船关键部位检测算法[J]. 系统工程与电子技术, 2026, 48(2): 410-421.

Yao WANG, Huiqi XU, Silei CAO, Lei WANG. Detection algorithm of ship critical parts based on deep learning[J]. Systems Engineering and Electronics, 2026, 48(2): 410-421.

图/表 20

图1

表1

图2

图3

图4

图5

图6

表2

表3

图7

表4

图8

表5

图9

表6

图10

表7

表8

表9

图11

参考文献 41

1	路建功, 徐国勇. 以科技视角审视未来战争形态[M]. 北京: 中国宇航出版社, 2022.
	LU J G, XU G Y. Examine the future forms of warfare from a technological perspective [M]. Beijing: China Astronautic Publishing House, 2022.
2	DING J, DAFOE A. Engines of power: electricity, AI, and general-purpose, military transformations[J]. European Journal of International Security, 2023, 8 (3): 377- 394. doi: 10.1017/eis.2023.1
3	张凯. 海战场环境下反舰导弹效能评估方法研究[J]. 舰船电子工程, 2024, 44 (6): 133- 136. doi: 10.3969/j.issn.1672-9730.2024.06.028
	ZHANG K. Research on effectiveness evaluation method for anti-ship missile in marine battlefield environment[J]. Ship Electronic Engineering, 2024, 44 (6): 133- 136. doi: 10.3969/j.issn.1672-9730.2024.06.028
4	苑桂萍, 肖益, 韩丛英. 2023年国外飞航导弹发展综述[J]. 战术导弹技术, 2024, 4 (2): 1- 8. doi: 10.16358/j.issn.1009-1300.20240508
	YUAN G P, XIAO Y, HAN C Y. A summary of development of foreign cruise missiles in 2023[J]. Tactical Missile Technology, 2024, 4 (2): 1- 8. doi: 10.16358/j.issn.1009-1300.20240508
5	张弘森, 徐建, 吴飞, 等. 基于深度学习的视觉图像舰船目标检测与识别方法综述[J]. 指挥信息系统与技术, 2024, 15 (6): 48- 63. doi: 10.15908/j.cnki.cist.2024.06.007
	ZHANG H S, XU J, WU F, et al. Review of ship target detection and recognition methods with visual imagery based on deep learning[J]. Command Information System and Technology, 2024, 15 (6): 48- 63. doi: 10.15908/j.cnki.cist.2024.06.007
6	高兵. 基于深度学习的舰船及其关键部位检测算法 [D]. 长春: 吉林大学, 2024.
	GAO B. Ship and vital parts of detection algorithms based on deep learning [D]. Changchun: Jilin University, 2024.
7	ZHANG H X, YU H S, TAO Y D, et al. Improvement of ship target detection algorithm for YOLOv7-tiny[J]. IET Image Processing, 2024, 18 (7): 1710- 1718.
8	JIA X Y, WANG H Q, YA Y D, et al. Anchor free SAR image ship target detection method based on the YOLO framework[J]. Systems Engineering & Electronics, 2022, 44 (12): 3703.
9	李双伟. 基于红外偏振成像的近岸舰船目标识别和干扰技术研究 [D]. 烟台: 烟台大学, 2016.
	LI S W. Research on inshore ship target recognition and antijamming technology based on infrared polarization imaging [D]. Yantai: Yantai University, 2016.
10	乔延军. 面向户外增强现实的地理实体目标识别与检测 [D]. 武汉: 武汉大学, 2017.
	QIAO Y J. Object recognition and detection of geographical entity based on outdoor augmented reality [D]. Wuhan: Wuhan University, 2017.
11	HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786):504−507.
12	HE K M, ZHAGN X Y, REN S Q, et al. Deep residual learning for image recognition [C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770−778.
13	苏娟, 杨龙, 黄华, 等. 用于SAR图像小目标舰船检测的改进SSD算法[J]. 系统工程与电子技术, 2020, 42 (5): 1026- 1034. doi: 10.3969/j.issn.1001-506X.2020.05.08
	SU J, YANG L, HUANG H, et al. Improved SSD algorithm for small-sized SAR ship detection[J]. Systems Engineering and Electronics, 2020, 42 (5): 1026- 1034. doi: 10.3969/j.issn.1001-506X.2020.05.08
14	GAO F, HE Y S, WANG J, et al. Anchor-free convolutional network with dense attention feature aggregation for ship detection in SAR images[J]. Remote Sensing, 2020, 12 (16): 2619. doi: 10.3390/rs12162619
15	白玉, 迟文恺, 谢宝蓉, 等. 结合目标提取和深度学习的红外舰船检测[J]. 电讯技术, 2023, 63 (2): 193- 198. doi: 10.20079/j.issn.1001-893x.211111002
	BAI Y, CHI W K, XIE B R, et al. Infrared ship detection combined with target extraction and deep learning[J]. Telecommunication Technology, 2023, 63 (2): 193- 198. doi: 10.20079/j.issn.1001-893x.211111002
16	TANG G, LIU S B, FUJINO I, et al. H-YOLO: a single-shot ship detection approach based on region of interest preselected network[J]. Remote Sensing, 2020, 12 (24): 4192. doi: 10.3390/rs12244192
17	刘忻伟, 朴永杰, 郑亮亮, 等. 面向航天光学遥感复杂场景图像的舰船检测[J]. 光学精密工程, 2023, 31 (6): 892- 904. doi: 10.37188/OPE.20233106.0892
	LIU X W, PU Y J, ZHENG L L, et al. Ship detection for complex scene images of space optical remote sensing[J]. Optical and Precision Engineering, 2023, 31 (6): 892- 904. doi: 10.37188/OPE.20233106.0892
18	林鑫伟, 徐志京, 黄海. 复杂背景下的SAR图像多尺度舰船检测[J]. 中国航海, 2023, 46 (2): 17- 24，32. doi: 10.3969/j.issn.1000-4653.2023.02.003
	LIN X W, XU Z J, HUANG H. Multi-scale ship detection in SAR image with complex background[J]. Navigation of China, 2023, 46 (2): 17- 24，32. doi: 10.3969/j.issn.1000-4653.2023.02.003
19	王玺坤, 姜宏旭, 林珂玉. 基于改进型YOLO算法的遥感图像舰船检测[J]. 北京航空航天大学学报, 2020, 46 (6): 1184- 1191.
	WANG X K, JIANG H X, LIN K Y. Remote sensing image ship detection based on modified YOLO algorithm[J]. Journal of Beijing University of Aeronautics and Astronautics, 2020, 46 (6): 1184- 1191.
20	LI Y S, LIU L Z, LU T W. SAE-CenterNet: self-attention enhanced CenterNet for small dense object detection[J]. Electronics Letters （Wiley-Blackwell）, 2023, 59 (3): 12732.
21	ZHOU X Y, WANG D Q, PHILIPP K, et al. Objects as points [EB/OL]. [2024-10-20]. https://arxiv.org/abs/1904.07850.
22	LIN H, JEGELKA S. ResNet with one-neuron hidden layers is a universal approximator [C]//Proc. of the Conference and Workshop on Neural Information Processing Systems, 2018: 6172−6181.
23	YU F, WANG D Q, SHELHAMER E, et al. Deep layer aggregation [C]//Proc. of the IEEE/CVF conference Computer Vision and Pattern Recognition, 2018: 2403−2413.
24	ZHOU X Y, WANG D Q, KRHENBUHL P. CenterNet: objects as points [EB/OL]. [2024-10-21]. https://arXiv.org/abs/1904.07850v1.
25	YU F, WANG D Q, SHELHAMER E, DARRELL T. Deep layer aggregation [C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 2403−2412.
26	LIU S T, HUANG D, WANG Y H. Receptive field block net for accurate and fast object detection[C]// Proc. of the European Conference on Computer Vision, 2018: 385−400.
27	李晨瑄, 顾佼佼, 王磊等. 多尺度特征融合的Anchor-Free轻量化舰船关键部位检测算法[J]. 北京航空航天大学学报, 2022, 48 (10): 2006- 2019. doi: 10.13700/j.bh.1001-5965.2021.0050
	LI C X, GU J J, WANG L, et al. Warship’s vital parts detection algorithm based on lightweight anchor-free network with multi-scale feature fusion[J]. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48 (10): 2006- 2019. doi: 10.13700/j.bh.1001-5965.2021.0050
28	YANG L, ZHANG R Y, LI L, et al. SimAM: a simple, parameter-free attention module for convolutional neural networks [C]// Proc. of the Machine Learning Research, 2021: 11863−11874.
29	HU J, SHEN L, SUN G, et al. Squeeze-and-excitation networks[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011−2023.
30	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module [C]// Proc. of the European Conference on Computer Vision, 2018: 3−19.
31	WANG Q L, WU B G, ZHU P F, et al. ECA−Net: efficient channel attention for deep convolutional neural networks [C]// Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 11531−11539.
32	LING X Y, ZHANG R Y, XIE X H. SimAM: a simple, prameter-free attention module for convolutional neural networks [C]// Proc. of the International Conference on Machine Learning, 2021: 1173−1198.
33	ZHU X F, WANG Q M, ZHANG B F, et al. An improved feature enhancement CenterNet model for small object defect detection on metal surfaces[J]. Advanced Theory & Simulations, 2024, 7 (8): 2301230.
34	AI-FURAIJI O J, TUAN N A, TSVIATKOU V Y. A new fast efficient non-maximum suppression algorithm based on image segmentation[J]. Indonesian Journal of Electrical Engineering and Computer Science, 2020, 19 (2): 1062- 1070. doi: 10.11591/ijeecs.v19.i2.pp1062-1070
35	SONG S B, ZHONG F, WANG T J, et al. Guided linear upsampling[J]. ACM Trans. on Graphics, 2023, 42 (4): 3592453.
36	CHENG T H, WANG X G, HUANG L C, et al. Boundary-preserving mask R-CNN [C]// Proc. of the European Conference on Computer Vision, 2020: 660−676.
37	LIU S, SUN Y, ZHANG L Y, et al. Fault diagnosis of shipboard medium-voltage alternating current power system with fault recording data-driven SE-ResNet18-1 model[J]. IEEJ Transactions on Electrical and Electronic Engineering, 2021, 124 (1): 106399.
38	LI W H, LIU M Y, LIU H, et al. Hourglass Tokenizer for efficient transformer-based 3D human pose estimation [C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024: 604−613.
39	XU L J, WANG Y H, SHI X S, et al. Real-time and accurate detection of citrus in complex scenes based on HPL-YOLOv4[J]. Computers and Electronics in Agriculture, 2023, 205, 107590. doi: 10.1016/j.compag.2022.107590
40	LI W, . SOLIHIN M I, NUGROHO H A. RCA: YOLOv8-based surface defects detection on the inner wall of cylindrical high-precision parts[J]. Arabian Journal for Science and Engineering, 2024, 49 (9): 12771- 12789. doi: 10.1007/s13369-023-08483-4
41	PURKAIT P, ZHAO C, ZACH C. SPP-Net: deep absolute pose regression with synthetic views [C]// Proc. Of Computer Vision and Pattern Recognition, 2018: 1101−1256.

骨干网络	AP/%	FPS
ResNet-18	28.1	142
DLA-34	37.4	52
Hourglass-104	45.1	1.4

注意力机制	参数量
SE^[29]	$ 2{C}^{{{}^{2}}}/r $
CBAM^[30]	$ 2{C}^{{{}^{2}}}/r+2{k}^{2} $
ECA^[31]	$ k $
SimAM	0

模型	Top-1- 准确率/%	Top-5-准确率/%	参数量/%	参数增加量/（×10⁶）	每秒浮点运算次数（十亿次/秒）	FPS
ResNet-18	70.33	89.58	11.69	0	1.82	215
+ SE	71.19	90.21	11.78	0.087	1.82	144
+ CBAM	71.24	90.04	11.78	0.090	1.82	78
+ ECA	70.71	89.85	11.69	36	1.82	148
+ SimAM	71.31	89.88	11.69	0	1.82	147
ResNet-34	73.75	91.60	21.80	0	3.67	119
+ SE	74.32	91.99	21.95	0.157	3.67	81
+ CBAM	74.41	91.85	21.96	0.163	3.67	38
+ ECA	74.03	91.73	21.81	74	3.67	82
+ SimAM	74.46	92.02	21.80	0	3.67	78
ResNet-101	77.82	93.85	44.55	0	7.83	47
+ SE	78.39	94.13	49.29	4.743	7.85	33
+ CBAM	78.57	94.18	49.33	4.781	7.85	14
+ ECA	78.46	94.12	44.55	171	7.84	33
+ SimAM	78.65	94.11	44.55	0	7.83	32
MobileNetV2	71.90	90.51	3.50	0	0.31	99
+ SE	72.46	90.85	3.53	0.028	0.31	65
+ CBAM	72.49	90.78	3.54	0.032	0.32	35
+ ECA	72.01	90.46	3.50	59	0.31	66
+ SimAM	72.36	90.74	3.50	0	0.31	66

关键部位	数量
驾驶舱	6945
天线桅杆	7197
雷达	3237
水线	8158

参数	取值
迭代次数	200
学习率	1.25×10⁻⁴
输入图像尺寸	512×512
批训练规模	8