复杂背景SAR图像舰船目标检测方法

doi:10.12305/j.issn.1001-506X.2026.10.15

摘要/Abstract

摘要：

针对复杂背景合成孔径雷达图像舰船目标检测精度低、易受干扰的问题，提出一种融合注意力机制与全局信息的SAR舰船目标检测方法。首先，构建感受野增强空间金字塔池化模块，缓解单一池化种类对数据处理不充分的问题，为后续检测环节提供更加精准的判断依据。然后，设计多重下采样模块，通过多分支特征提取实现在下采样过程中保留更多细节，增强算法对舰船目标的感知能力。最后，使用空间与通道重建卷积优化颈部网络，利用其重构能力有效抑制背景干扰。实验结果表明，所提算法相较于基线算法，在所选数据集上平均精度（交并比阈值为0.5）分别提高2.2%和2.5%，达到94.5%和89.1%，同时算法的浮点运算数降低9.5%。在保持较小模型规模的前提下，能够有效提升复杂背景下合成孔径雷达舰船目标的检测性能。

关键词: 舰船目标检测, YOLOv11, 多重下采样, 空间金字塔池化, 空间与通道重建卷积

Abstract:

To address the issue of low accuracy and susceptibility to interference in synthetic aperture radar (SAR) ship images detection with complex backgrounds, a ship target detection method integrating attention mechanism and global information is proposed. First, a receptive field enhanced spatial pyramid pooling module is constructed to alleviate the problem of insufficient feature extraction by a single pooling type, providing a more accurate basis for subsequent detection. Second, a multi-scale downsampling module is designed to retain more details during downsampling through multi-branch feature extraction, enhancing the algorithm’s perception of ship targets. Finally, spatial and channel reconstruction convolution is used to optimize the neck network, effectively suppressing background interference with its reconstruction capability. Experimental results show that the proposed algorithm outperforms the baseline algorithm by 2.2% and 2.5% in average precision (with an intersection over union threshold of 0.5) selected datasets, reaching 94.5% and 89.1% respectively, while reducing the number of floating point operations per second by 9.5%. Under the premise of maintaining a compact model size, the proposed method effectively improves the detection performance of SAR ship targets in complex backgrounds.

Key words: ship target detection, YOLOv11, multiple downsampling, spatial pyramid pooling, spatial and channel reconstruction convolution

中图分类号:

TP 391.4

赵晶钰, 李敏, 陈谢发, 杨爱涛, 何玉杰. 复杂背景SAR图像舰船目标检测方法[J]. 系统工程与电子技术, 2026, 48(10): 3427-3437.

Jingyu Zhao, Min Li, Xiefa Chen, Aitao Yang, Yujie He. Ship target detection method for SAR images with complex backgrounds[J]. Systems Engineering and Electronics, 2026, 48(10): 3427-3437.

图/表 16

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

表1

表2

表3

表4

表5

图11

参考文献 30

1	邓云凯, 禹卫东, 张衡, 等. 未来星载SAR技术发展趋势[J]. 雷达学报, 2020, 9 (1): 1- 3.
2	Wu X, He X, Tian S J, et al. Review of ship target detection based on SAR images [C]//7th International Conference on Transportation Information and Safety, 2023: 2106.
3	Asiyabi R M, Ghorbanian A, Tameh S N, et al. Synthetic aperture radar （SAR） for ocean: a review[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 16, 9106. doi: 10.1109/JSTARS.2023.3310363
4	Li J W, Chen J, Cheng P, et al. A survey on deep-learning-based real-time SAR ship detection[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 16, 3218. doi: 10.1109/JSTARS.2023.3244616
5	Muhammad Y, Wan J H, Xu M M, et al. Ship detection based on deep learning using SAR imagery: a systematic literature review[J]. Soft Computing, 2023, 27 (1): 63. doi: 10.1007/s00500-022-07522-w
6	Liu Q K, Liu R, Zheng B L, et al. Infrared small target detection with scale and location sensitivity [C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024: 17490.
7	Moreira A, Prats-Iraola P, Younis M, et al. A tutorial on synthetic aperture radar[J]. IEEE Geoscience and Remote Sensing Magazine, 2013, 1 (1): 6. doi: 10.1109/MGRS.2013.2248301
8	王浩宇, 杨海涛, 王晋宇, 等. 遥感图像去噪方法研究综述[J]. 计算机工程与应用, 2024, 60 (15): 55.
9	Gong Y C, Zhang Z, Wen J B, et al. Small ship detection of SAR images based on optimized feature pyramid and sample augmentation[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 16, 7385. doi: 10.1109/JSTARS.2023.3302575
10	Huang Q N, Zhu W G, Li Y G, et al. Survey of target detection algorithms in SAR images [C]//IEEE 5th Advanced Information Technology, Electronic and Automation Control Conference, 2021: 1756.
11	Xing X W, Chen Z L, Zou H X, et al. A fast algorithm based on two-stage CFAR for detecting ships in SAR images [C]//Asian-Pacific Conference on Synthetic Aperture Radar, 2009: 506.
12	Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation [C]//IEEE Conference on Computer Vision and Pattern Recognition, 2014: 580.
13	Ren S Q, He K M, Girshick R, et al. Faster R-CNN: towards real-time object detection with region propos-al networks[J]. IEEE Trans. on Pattern Analysis & Machine Intelligence, 2017, 39 (6): 1137.
14	Wang C Y, Bochkovskiy A, Liao H Y. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for realtime object detectors [C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 7464.
15	Wang C Y, Yeh I H, Liao H Y. Yolov9: learning what you want to learn using programmable gradient information [C]//European Conference on Computer Vision, 2024.
16	Wang A, Chen H, Liu L H, et al. Yolov10: real-time end-to-end object detection[J]. Advances in Neural Information Processing Systems, 2024, 37, 107984.
17	Liu W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector [C]//European Conference on Computer Vision, 2016: 21.
18	Pan D C, Gao X, Dai W, et al. SRT-Net: scattering region topology network for oriented ship detection in large-scale SAR images[J]. IEEE Trans. on Geoscience and Remote Sensing, 2024, 62
19	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. doi: 10.1080/01431161.2023.2173030
20	Tang X, Zhang J F, Xia Y Z, et al. DBW-YOLO: a high-precision SAR ship detection method for complex environments[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17, 7029. doi: 10.1109/JSTARS.2024.3376558
21	杨明秋, 左小清, 董燕. 基于改进YOLOv8s的SAR图像舰船目标检测方法[J]. 激光与光电子学进展, 2024, 61 (22): 306.
22	Li J F, Wen Y, He L H. SCConv: spatial and channel reconstruction convolution for feature redundancy [C]//IEEE/ CVF Conference on Computer Vision and Pattern Recognition, 2023: 6153.
23	Khanam R, Hussain M. Yolov11: an overview of the key architectural enhancements[EB/OL]. [2025-06-05]. https://doi.org/10.48550/arXiv.2410.17725.
24	Zhang X, Liu C, Yang D G, et al. RFAConv: innovating spatial attention and standard convolutional operation[EB/OL]. [2025-06-05]. https://doi.org/10.48550/arXiv.2304.03198.
25	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
26	Wei S J, Zeng X F, Qu Q Z, et al. HRSID: a high-resolution SAR images dataset for ship detection and instance segmentation[J]. IEEE Access, 2020, 8, 120234. doi: 10.1109/ACCESS.2020.3005861
27	Padilla R, Netto S L, SILVA E. A survey on performance metrics for object-detection algorithms [C]//International Conference on Systems, Signals and Image Processing, 2020: 237.
28	Shao Y H, Zhang D, Chu H Y, et al. A review of YOLO object detection based on deep learning[J]. Journal of Electronics and Information Technology, 2022, 44 (10): 3697.
29	Li C Y, Li L L, Jiang H L, et al. YOLOv6: a single-stage object detection framework for industrial applications[EB/OL]. [2025-06-05]. https://doi.org/10.48550/arXiv.2209.02976.
30	Zhao Y A, Lv W Y, Xu S L, et al. DETRS beat YOLOs on real-time object detection [C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024: 16965.

参数说明	参数值
迭代批次样本数量	4
工作进程数	32
训练次数	100
动量因子	0.9
初始学习率	0.01

数据集	算法	多重下采样	感受野增强金字塔池化结构	空间与通道重建卷积	P/%	R/%	AP₅₀/%	AP_50:95/%	Par（×10⁶）	GFLOPS
SSDD	基线算法				92.1	86.9	92.3	60.6	2.58	6.3
	①	√			92.3	87.1	93.0	62.7	2.31	5.4
	②		√		94.7	81.9	93.0	61.3	2.88	6.4
	③			√	89.6	87.7	92.7	61.9	2.69	6.5
	④	√	√		94.0	86.8	93.8	62.8	2.61	5.4
	⑤	√		√	93.3	86.9	94.1	63.0	2.42	5.6
	⑥		√	√	91.7	88.6	94.2	62.0	2.99	6.6
	⑦	√	√	√	95.2	88.8	94.5	62.1	2.72	5.7
HRSID	基线算法				88.1	77.9	86.9	61.0	2.58	6.3
	①	√			89.8	78.5	88.0	61.7	2.31	5.6
	②		√		90.2	78.8	87.8	60.9	2.88	6.4
	③			√	89.1	78.4	87.7	61.3	2.69	6.5
	④	√	√		91.0	78.1	88.4	61.4	2.61	5.4
	⑤	√		√	89.1	80.3	88.7	62.5	2.42	5.6
	⑥		√	√	91.0	77.8	88.5	62.5	2.99	6.6
	⑦	√	√	√	91.5	78.0	89.1	62.7	2.72	5.7

模型	P/%	R/%	AP₅₀/%	AP_50:95/%	Par（×10⁶）	GFLOPS	FPS
Faster R-CNN	79.6	88.8	87.6	52.1	28.68	138.3	26.4
SSD	94.1	83.5	91.5	51.0	23.74	30.4	56.7
YOLOv5n	88.4	88.9	93.8	62.4	2.50	7.1	644.45
YOLOv6n	93.0	86.4	93.0	62.5	4.23	11.8	865.68
YOLOv7tiny	90.5	86.5	93.2	62.2	6.01	13.2	782.86
YOLOv8n	93.9	86.6	93.3	62.6	3.00	8.1	773.09
YOLOv9t	90.5	86.5	93.2	62.2	1.97	7.6	285.66
YOLOv10n	81.6	80.1	86.6	57.9	2.69	8.2	920.64
YOLOv11n	92.1	86.9	92.3	60.6	2.58	6.3	799.82
RT-DETR-r18	90.8	81.2	90.1	62.8	19.82	57.2	220.95
CB-YOLO	95.2	86.8	94.5	63.1	2.72	5.7	651.46

模型	P/%	R/%	AP₅₀/%	AP_50:95/%	Par（×10⁶）	GFLOPS	FPS
Faster R-CNN	71.5	81.9	81.0	55.8	28.68	102	28.1
SSD	80.5	77.9	79.1	50.2	23.74	30.4	70.6
YOLOv5n	89.4	79.9	88.5	62.6	2.50	7.1	682.49
YOLOv6n	87.9	77.1	85.4	59.0	4.23	11.8	711.38
YOLOv7tiny	89.7	76.7	86.0	55.8	6.01	13.2	756.36
YOLOv8n	89.9	80.4	89.0	63.1	3.00	8.1	684.97
YOLOv9t	89.2	80.8	88.9	63.0	1.97	7.6	517.93
YOLOv10n	87.1	77.6	86.2	60.5	2.69	8.2	691.07
YOLOv11n	88.1	77.9	86.9	61.0	2.58	6.3	660.77
RT-DETR-r18	90.4	81.3	89.7	48.9	19.88	57.2	244.12
CB-YOLO（ours）	91.5	78.0	89.1	63.3	2.72	5.7	558.74

是否加载预训练权重	P/%	R/%	AP₅₀/%	AP_50:95/%	Par（×10⁶）	GFLOPS
否	0.952	0.868	0.945	0.621	2.72	5.7
是	0.928	0.919	0.952	0.642	2.72	5.7