混合注意力优化的SAR图像小目标检测方法

doi:10.12305/j.issn.1001-506X.2025.08.10

摘要/Abstract

摘要：

近年来，卷积神经网络在合成孔径雷达（synthetic aperture radar, SAR）图像船舶检测中取得突出成就，但小目标检测方面仍然存在较大不足。对此，提出一种基于YOLO （you only look once） v5的改进检测网络，结合空间感知通道注意力、自注意力机制和上下文特征融合策略，以提高小型船舶的检测性能。首先，通道注意力机制抑制了背景信息并强调目标特征，显著提高检测精度。其次，在YOLOv5的骨干网络和检测层中引入自注意力模块，以捕获全局信息，增强定位能力。最后，通过融合浅层和深层特征，补充特征提取中丢失的小目标信息，进一步提高检测精度。基于大规模SAR船舶监测数据集（large-scale SAR ship detection dataset version 1.0 LS-SSDDv1.0）数据集的实验结果表明，改进后的网络的全类平均精度(mean average precision，mAP)0.5指标达78.9%，显著优于现有方法。

关键词: 合成孔径雷达图像, 船舶检测, 注意力机制, 特征融合, 小目标检测

Abstract:

In recent years, convolutional neural network （CNN） achieves remarkable success in synthetic aperture radar （SAR） image ship detection. However, there are still considerable challenges in detecting small targets. Regarding this, an improved detection network based on you only look once （YOLO） v5, combining the spatial-aware channel attention （SCA）, self-attention mechanism, and contextual feature fusion （CFF） strategy to enhance the detection performance of small ships. Firstly, SCA improves detection accuracy by suppressing background information and highlighting target-related features. Secondly, self-attention module is introduced in the backbone network and detection lager of YOLOv5 to capture global information and enhance localization capabilities. Finally, by fusing shallow and deep features, the compensates for the loss of small target information during feature extraction, further improving detection accuracy. Experimental results based on the LS-SSDDv1.0 dataset show that the improved network achieves a mean average precision （mAP） 0.5 of 78.9%, significantly outperforming existing methods.

Key words: synthetic aperture radar (SAR) image, ship detection, attention mechanism, feature fusion, small target detection

中图分类号:

TP 391.4

付卫红, 彭文洪, 刘乃安. 混合注意力优化的SAR图像小目标检测方法[J]. 系统工程与电子技术, 2025, 47(8): 2519-2526.

Weihong FU, Wenhong PENG, Naian LIU. SAR image small target detection method with hybrid attention optimization[J]. Systems Engineering and Electronics, 2025, 47(8): 2519-2526.

图/表 12

图1

图2

图3

图4

表1

表2

表3

表4

表5

图5

图6

表6

参考文献 30

1	ZHANG T W, ZHANG X L, LI J, et al. SAR ship detection dataset （SSDD）: official release and comprehensive data analysis[J]. Remote Sensing, 2021, 13 (18): 3690. doi: 10.3390/rs13183690
2	DI-BISCEGLIE M, GALDI C. CFAR detection of extended objects in high-resolution SAR images[J]. IEEE Trans. on Geoscience and Remote Sensing, 2005, 43 (4): 833- 843. doi: 10.1109/TGRS.2004.843190
3	XU M H, ZHU J, FANG J, et al. CFAR based NOMP for line spectral estimation and detection[J]. IEEE Trans. on Aerospace and Electronic Systems, 2023, 59 (5): 6971- 6990.
4	RIHAN M Y, NOSSAIR Z B, MUBARAK R I. An improved CFAR algorithm for multiple environmental conditions[J]. Signal, Image and Video Processing, 2024, 18 (4): 3383- 3393. doi: 10.1007/s11760-024-03001-x
5	宋志娜, 眭海刚, 李永成. 高分辨率可见光遥感图像舰船目标检测综述[J]. 武汉大学学报（信息科学版）, 2021, 46 (11): 1703- 1715.
	SONG Z N, SUI H G, LI Y C. A review of ship target detection in high-resolution visible light remote sensing images[J]. Geomatics and Information Science of Wuhan University, 2021, 46 (11): 1703- 1715.
6	XU D Q, WU Y Q. Research progress of deep learning algorithms for optical remote sensing image target detection[J]. Journal of Remote Sensing, 2024, 28 (12): 3405.
7	GUO H Y, YANG X, WANG N N, et al. A CenterNet++ model for ship detection in SAR images[J]. Pattern Recognition, 2021, 112, 107787. doi: 10.1016/j.patcog.2020.107787
8	LI D, LIANG Q H, LIU H Q, et al. A novel multidimensional domain deep learning network for SAR ship detection[J]. IEEE Trans. on Geoscience and Remote Sensing, 2022, 60, 5203213.
9	JIANG Z J, WANG Y P, ZHOU X, et al. Small-scale ship detection for SAR remote sensing images based on coordinate-aware mixed attention and spatial semantic joint context[J]. Smart Cities, 2023, 6 (3): 1612- 1629. doi: 10.3390/smartcities6030076
10	GAO S, LIU J M, MIAO Y H, et al. A high-effective implementation of ship detector for SAR images[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19, 4019005.
11	FANG H Z, LIAO Z, WANG X H, et al. Differentiated attention guided network over hierarchical and aggregated features for intelligent UAV surveillance[J]. IEEE Trans. on Industrial Informatics, 2023, 19 (9): 9909- 9920. doi: 10.1109/TII.2022.3232777
12	VASWANI A, SHAZEER N, PARMAR N. Attention is all you need[C]//Proc. of the 31st International Conference on Neural Information Processing Systems, 2017: 6000−6010.
13	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: 936−944.
14	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.
15	ZHU X K, LYU S C, WANG X, et al. TPH-YOLOv5: improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]//Proc. of the IEEE/CVF International Conference on Computer Vision Workshops, 2021: 2778−2788.
16	HU J, SHEN L, ALBANIE S, et al. Squeeze-and-excitation networks[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2020, 42 (8): 2011- 2023. doi: 10.1109/TPAMI.2019.2913372
17	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[M]. Cham, Switzerland: Springer, 2018.
18	NIU Z Y, ZHONG G Q, YU H. A review on the attention mechanism of deep learning[J]. Neurocomputing, 2021, 452, 48- 62. doi: 10.1016/j.neucom.2021.03.091
19	HAN K, WANG Y H, CHEN H T, et al. A survey on vision transformer[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2022, 45 (1): 87- 110.
20	LIU Z, NING J, CAO Y, et al. Video swin transformer[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 3202−3211.
21	DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16×16 words: transformers for image recognition at scale[EB/OL]. [2024-05-03]. https://arxiv.org/abs/2010.11929.
22	ZHANG Y, ZHU G Y, SHI T J, et al. Remote sensing image small target detection based on feature fusion and attention[J]. Acta Optica Sinica, 2022, 42 (24): 2415001. doi: 10.3788/AOS202242.2415001
23	ZHANG T W, ZHANG X L, KE X, et al. LS-SSDD-v1.0: a deep learning dataset dedicated to small ship detection from large-scale Sentinel-1 SAR images[J]. Remote Sensing, 2020, 12 (18): 2997. doi: 10.3390/rs12182997
24	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- 3708.
25	REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2017, 39 (6): 1137- 1149. doi: 10.1109/TPAMI.2016.2577031
26	WU Y, CHEN Y P, YUAN L, et al. Rethinking classification and localization for object detection[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 10183−10192.
27	SU N, HE J Y, YAN Y M, et al. SII-Net: spatial information integration network for small target detection in SAR images[J]. Remote Sensing, 2022, 14 (3): 442. doi: 10.3390/rs14030442
28	BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection [EB/OL]. [2024-05-03]. https://arxiv.org/abs/2004.10934.
29	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023: 7464−7475.
30	GE Z, LIU S T, WANG F, et al. YOLOX: exceeding YOLO series in 2021[EB/OL]. [2024-05-03]. http://arxiv.org/abs/2107.08430.

SCA	C3TR	TPH	CON3	@AP0.5	@AP0.5:0.95
—	—	—	—	74.5	28.4
√	—	—	—	77.4	29.7

SCA	C3TR	TPH	CON3	@AP0.5	@AP0.5:0.95
—	—	—	—	74.5	28.4
—	√	—	—	76.7	28.6
—	—	√	—	78.0	29.1

SCA	C3TR	TPH	CON3	@AP0.5	@AP0.5:0.95
—	—	—	—	74.5	28.4
—	—	—	√	75.8	28.8

SCA	C3TR	TPH	CON3	@AP0.5	@AP0.5:0.95
—	—	—	—	74.5	28.4
√	√	—	—	77.7	29.4
√	—	√	—	78.6	29.2
√	√	—	√	78.3	29.7
√	√	√	√	78.9	30.0

网络结构	P	R	@AP0.5	@AP0.5:0.95
Faster R-CNN^[25]	73.8	77.7	74.8	—
Double-Head R-CNN^[26]	74.4	78.5	75.7	—
SII-Net^[27]	79.3	68.2	76.1	—
SAR-Net^[6]	58.4	82.0	76.2	—
YOLOv4^[28]	48.2	80.6	72.8	—
YOLOv5m	82.8	67.2	74.5	28.4
YOLOv7^[29]	81.0	64.1	71.2	26.7
YOLOX^[30]	66.0	77.4	74.0	—
YOLOv8m	83.5	66.7	76.0	30.6
YOLOX+CMA+SSJC^[9]	70.4	80.1	77.2	—
本文方法	85.5	71.4	78.9	30.0