基于全局位置信息和残差特征融合的SAR船舶检测算法

doi:10.12305/j.issn.1001-506X.2024.03.09

摘要/Abstract

摘要：

针对合成孔径雷达(synthetic aperture radar, SAR)图像船舶目标尺度不一且易受海面、地面杂波和相干斑噪声的影响, 难以提取目标多维特征且特征融合过程中易产生语义歧义, 造成船舶目标检测率低, 虚警率高的问题, 提出一个基于全局位置信息和残差特征融合的SAR船舶目标检测算法。基于Faster区域卷积神经网络(region convolutional neural network, R-CNN)目标检测算法, 在特征提取网络和特征融合网络中进行改进: 在特征提取网络中使用高宽注意力机制提取目标在图像中的全局位置信息, 增强目标的多维特征提取能力; 在特征融合网络中使用带有残差连接的双向特征金字塔网络削弱特征融合过程中的语义歧义, 降低复杂背景下的船舶目标虚警率, 同时进行不同层级的多尺度特征双向融合, 增强高低层特征的联系, 提升多尺度船舶目标的检测能力。在SAR船舶数据集上达到98.2%的均值平均精度, 超过部分算法2.4%以上。实验表明, 所提算法有效提取了目标的多维特征, 显著缓解了语义歧义问题, 具有较好的检测能力和泛化能力。

关键词: 合成孔径雷达, 船舶检测, 注意力机制, 特征金字塔网络, 残差连接

Abstract:

In view of the problem that ship targets in synthetic aperture radar (SAR) images have varying scales and are easily affected by sea clutter, ground clutter, and coherent speckle noise, which makes it difficult to extract target multidimensional features and leads to semantic ambiguity during feature fusion, resulting in low ship target detection rate and high false alarm rate, a SAR ship target detection algorithm based on global position information and residual feature fusion is proposed. Based on the Faster region convolutional neural network (R-CNN) object detection algorithm, improvements are made in the feature extraction network and feature fusion network. A height-width attention mechanism is used in the feature extraction network to extract the global position information of the target in the image, enhancing the multidimensional feature extraction capability of the target. A bidirectional feature pyramid network with residual connections is used in the feature fusion network to reduce semantic ambiguity in the feature fusion process, reduce false alarms of ship targets in complex backgrounds, and perform bidirectional fusion of multi-scale features at different levels to enhance the connection between high and low-level features and improve the detection capability of multi-scale ship targets. The algorithm achieves a mean average precision of 98.2% on the SAR ship dataset, surpassing some algorithms by 2.4% or more. The experiments show that the proposed algorithm effectively extracts multidimensional features of the target, significantly alleviates semantic ambiguity problems, and has good detection and generalization capabilities.

Key words: synthetic aperture radar (SAR), ship detection, attention mechanism, feature pyramid network, residual connection

中图分类号:

TN95

方小宇, 黄丽佳. 基于全局位置信息和残差特征融合的SAR船舶检测算法[J]. 系统工程与电子技术, 2024, 46(3): 839-848.

Xiaoyu FANG, Lijia HUANG. SAR ship detection algorithm based on global position information and fusion of residual feature[J]. Systems Engineering and Electronics, 2024, 46(3): 839-848.

图/表 15

图1

图2

图3

图4

表1

表2

图5

图6

表3

图7

图8

表4

图9

表5

图10

参考文献 27

1	李健伟, 曲长文, 彭书娟, 等. 基于卷积神经网络的SAR图像舰船目标检测[J]. 系统工程与电子技术, 2018, 40 (9): 1953- 1959.
	LI J W , QU C W , PENG S J , et al. Ship detection in SAR images based on convolutional neural network[J]. Systems Engineering and Electronics, 2018, 40 (9): 1953- 1959.
2	MIAO T , ZENG H C , YANG W , et al. An improved lightweight retinaNet for ship detection in SAR images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2022, 15, 4667- 4679. doi: 10.1109/JSTARS.2022.3180159
3	RIZAEV I G , KARAKUŞ O , HOGAN S J , et al. Modeling and SAR imaging of the sea surface: a review of the state-of-the-art with simulations[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2022, 187, 120- 140. doi: 10.1016/j.isprsjprs.2022.02.017
4	YU J M , WU T , ZHOU S B , et al. An SAR ship object detection algorithm based on feature information efficient representation network[J]. Remote Sensing, 2022, 14 (14): 3489. doi: 10.3390/rs14143489
5	DENG Y W, GUAN D H, CHEN Y Y, et al. SAR-Shipnet: SAR-ship detection neural network via bidirectional coordinate attention and multi-resolution feature fusion[C]//Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing, 2022: 3973-3977.
6	LI W K, ZOU B, XIN Y, et al. An improved CFAR scheme for man-made target detection in high resolution SAR images[C]//Proc. of the IEEE International Geoscience and Remote Sensing Symposium, 2018: 2829-2832.
7	ZHU J W , QIU X L , PAN Z X , et al. Projection shape template-based ship target recognition in TerraSAR-X images[J]. IEEE Geoscience and Remote Sensing Letters, 2016, 14 (2): 222- 226.
8	YANG G Z , YU J , SUN W D . Ship wake detection in SAR images with complex backgrounds based on relative total variation[J]. Journal of University of Chinese Academy of Sciences, 2017, 34 (6): 734- 742.
9	GAO G , OUYANG K W , LUO Y B , et al. Scheme of parameter estimation for generalized gamma distribution and its application to ship detection in SAR images[J]. IEEE Trans.on Geoscience and Remote Sensing, 2016, 55 (3): 1812- 1832.
10	LENG X G , JI K F , YANG K , et al. A bilateral CFAR algorithm for ship detection in SAR images[J]. IEEE Geoscience and Remote Sensing Letters, 2015, 12 (7): 1536- 1540. doi: 10.1109/LGRS.2015.2412174
28	LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 2117-2125.
29	HOU Q B, ZHOU D Q, FENG J S. Coordinate attention for efficient mobile network design[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 13713-13722.
30	LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 8759-8768.
31	SZEGEDY C, LIU W, JIA Y Q, et al. Going deeper with convolutions[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2015.
32	ZHANG T W , ZHANG X L , LI J W , et al. SAR ship detection dataset (SSDD): official release and comprehensive data analysis[J]. Remote Sensing, 2021, 13 (18): 3690. doi: 10.3390/rs13183690
11	SHI Z W , YU X R , JIANG Z G , et al. Ship detection in high-resolution optical imagery based on anomaly detector and local shape feature[J]. IEEE Trans.on Geoscience and Remote Sensing, 2013, 52 (8): 4511- 4523.
12	HE K M, CHEN X L, XIE S N, et al. Masked autoencoders are scalable vision learners[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 16000-16009.
13	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2014: 580-587.
14	孙方伟, 李承阳, 谢永强, 等. 深度学习应用于遮挡目标检测算法综述[J]. 计算机科学与探索, 2022, 16 (6): 1243- 1259.
	SUN F W , LI C Y , XIE Y Q , et al. Review of deep learning applied to occluded object detection[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16 (6): 1243- 1259.
15	LAW H, DENG J. Cornernet: detecting objects as paired keypoints[C]//Proc. of the European Conference on Computer Vision, 2018: 734-750.
16	TIAN Z, SHEN C H, CHEN H, et al. FCOS: fully convolutional one-stage object detection[C]//Proc. of the IEEE/CVF International Conference on Computer Vision, 2019: 9627-9636.
17	周晓玲, 张朝霞, 鲁雅, 等. 基于改进R-FCN的SAR图像识别[J]. 系统工程与电子技术, 2022, 44 (4): 1202- 1209.
	ZHOU X L , ZHANG Z X , LU Y , et al. SAR image recognition based on improved R-FCN[J]. Systems Engineering and Electronics, 2022, 44 (4): 1202- 1209.
18	LI Y G , ZHU W G , LI C X , et al. SAR image near-shore ship target detection method in complex background[J]. International Journal of Remote Sensing, 2023, 44 (3): 924- 952. doi: 10.1080/01431161.2023.2173030
19	JIA X Y, WANG H Q, WANG M, et al. SAR image ship target detection based on sea-land segmentation and YOLO anchor free[C]//Proc. of the 14th International Conference on Machine Vision, 2022, 12084: 200-207.
20	杜兰, 王梓霖, 郭昱辰, 等. 结合强化学习自适应候选框挑选的SAR目标检测方法[J]. 雷达学报, 2022, 11 (5): 884- 896.
	DU L , WANG Z L , GUO Y C , et al. Adaptive region proposal selection for SAR target detection using reinforcement learning[J]. Journal of Radars, 2022, 11 (5): 884- 896.
21	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.
	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.
22	李丽圆, 李潇雁, 胡琸悦, 等. 基于回归模型与注意力的轻量化SAR舰船检测模型[J]. 红外与毫米波学报, 2022, 41 (3): 618- 625.
	LI L Y , LI X Y , HU Z Y , et al. The research on lightweight SAR ship detection method based on regression model and attention[J]. Journal of Infrared and Millimeter Waves, 2022, 41 (3): 618- 625.
23	郑子阳, 张婷, 刘兆英, 等. 基于旋转损失函数RCIOU的SAR图像舰船目标检测方法[J]. 山东大学学报(工学版), 2022, 52 (2): 15- 22.
	ZHENG Z Y , ZHANG T , LIU Z Y , et al. Ship target detection in SAR images based on RCIoU loss function[J]. Journal of Shandong University(Engineering Science), 2022, 52 (2): 15- 22.
24	韩子硕, 王春平, 付强. 基于深层次特征增强网络的SAR图像舰船检测[J]. 北京理工大学学报自然版, 2021, 41 (9): 1006- 1014.
	HAN Z S , WANG C P , FU Q . Ship detection in SAR images based on deep feature enhancement network[J]. Transactions of Beijing Institute of Technology, 2021, 41 (9): 1006- 1014.
25	ZHANG T W , ZHANG X L , KE X . Quad-FPN: a novel quad feature pyramid network for SAR ship detection[J]. Remote Sensing, 2021, 13 (14): 2771. doi: 10.3390/rs13142771
26	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7132-7141.
27	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.

α	mAP50/%
0(不添加残差连接)	93.6
0.25	95.3
0.5	96.5
1	95.6
2	91.5
自适应调整	不稳定

算法	mAP/%	mAP50/%	mAP75/%	小目标/%	中目标/%	大目标/%	速度/(iter/s)
Faster R-CNN	53.1	93.6	57.2	51.2	61.2	56.5	9.6
ImShips	50.4	93.1	54.3	51.4	54.7	58.6	15.5
QUAD-FPN	56.3	95.8	64.4	54.6	63.2	70.3	5.7
SSDNet	58.3	98.2	69.0	57.3	65.5	72.5	8.1

算法	mAP	mAP50	mAP75	小目标	中目标	大目标
Faster R-CNN	53.1	93.6	57.2	51.2	61.2	56.5
HWAM	58.3	97.0	65.0	55.5	63.0	70.0
ResFPN	54.7	96.5	61.7	56.2	62.3	75.1
HWAM+ResFPN	58.3	98.2	69.0	57.3	65.5	72.5

卷积结构	mAP50
3×3卷积	93.6
改良Inception	95.4

算法	mAP	mAP50	mAP75	小目标	中目标	大目标
Faster R-CNN	26.5	62.0	15.3	6.2	28.5	35.9
ImShips	32.4	68.5	26.6	8.7	35.3	39.5
QUAD-FPN	36.3	74.4	27.8	14.8	37.2	58.5
SSDNet	37.1	80.2	29.0	15.8	37.9	64.6