基于深度分离卷积神经网络的高速高精度SAR舰船检测

张晓玲; 张天文; 师君; 韦顺军

doi:10.12000/JR19111

基于深度分离卷积神经网络的高速高精度SAR舰船检测

DOI: 10.12000/JR19111

电子科技大学信息与通信工程学院成都 611731

基金项目: 国家自然科学基金(61571099, 61501098, 61671113)，国家重点研发计划(2017YFB0502700)

详细信息

作者简介:
张晓玲(1964–)，女，四川人，获电子科技大学工学博士学位，目前为电子科技大学教授/博导，主要从事SAR成像技术、雷达探测技术研究、3维SAR成像的目标散射特性(RCS)反演。E-mail: xlzhang@uestc.edu.cn

张天文(1994–)，男，江苏人，现于电子科技大学信息与通信工程学院攻读博士学位，主要研究领域为SAR成像技术、遥感图像处理与智能识别解译。E-mail: twzhang@std.usetc.edn.cn

师　君(1979–)，男，河南人，获电子科技大学工学博士学位，目前为电子科技大学副教授，主要从事SAR数据处理方面研究。E-mail: shijun@uestc.edu.cn

韦顺军(1983–)，男，广西人，获电子科技大学工学博士学位，目前为电子科技大学副教授，主要从事SAR成像技术、干涉SAR技术研究。E-mail: weishunjun@uestc.edu.cn

通讯作者:
张晓玲 xlzhang@uestc.edu.cn

中图分类号: TN957.52
计量
- 文章访问数: 7181
- HTML全文浏览量: 2339
- PDF下载量: 675
- 被引次数: 52
出版历程
- 收稿日期: 2019-12-16
- 修回日期: 2019-12-23
- 网络出版日期: 2019-12-01

High-speed and High-accurate SAR Ship Detection Based on a Depthwise Separable Convolution Neural Network

School of Electronic Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China

Funds: The National Natural Science Foundation of China (61571099, 61501098, 61671113), The National Key R&D Program of China (2017YFB0502700)

More Information

Corresponding author: ZHANG Xiaoling, xlzhang@uestc.edu.cn

摘要

摘要:
随着人工智能的兴起，利用深度学习技术实现SAR舰船检测，能够有效避免传统的复杂特征设计，并且检测精度获得了极大的改善。然而，现如今大多数检测模型往往以牺牲检测速度为代价来提高检测精度，限制了一些SAR实时性应用，如紧急军事部署、迅速海难救援、实时海洋环境监测等。为了解决这个问题，该文提出一种基于深度分离卷积神经网络(DS-CNN)的高速高精度SAR舰船检测方法SARShipNet-20，该方法取代传统卷积神经网络(T-CNN)，并结合通道注意力机制(CA)和空间注意力机制(SA)，能够同时实现高速和高精度的SAR舰船检测。该方法在实时性SAR应用领域具有一定的现实意义，并且其轻量级的模型有助于未来的FPGA或DSP的硬件移植。
- 卷积神经网络 /
- 深度分离卷积神经网络 /
- SAR /
- 舰船检测 /
- 注意力机制
Abstract:
With the development of artificial intelligence, Synthetic-Aperture Radar (SAR) ship detection using deep learning technology can effectively avoid traditionally complex feature design and thereby greatly improve detection accuracy. However, most existing detection models often improve detection accuracy at the expense of detection speed that limits some real-time applications of SAR such as emergency military deployment, rapid maritime rescue, and real-time marine environmental monitoring. To solve this problem, a high-speed and high-accuracy SAR ship detection method called SARShipNet-20 based on a Depthwise Separable Convolution Neural Network (DS-CNN) has been proposed in this paper, that replaces the Traditional Convolution Neural Network (T-CNN) and combines Channel Attention (CA) and Spatial Attention (SA). As a result, high-speed and high-accuracy SAR ship detection can be simultaneously achieved. This method has certain practical significance in the field of real-time SAR application, and its lightweight model is helpful for future FPGA or DSP hardware transplantation.
- Convolution Neural Network (CNN) /
- Depthwise Separable Convolution Neural Network (DS-CNN) /
- SAR /
- Ship detection /
- Attention mechanism

HTML全文

图 1 传统卷积神经网络和深度分离卷积神经网络示意图

Figure 1. Diagrammatic sketch of T-CNN and DS-CNN

下载: 全尺寸图片幻灯片

图 2 网络结构示意图 (SARShipNet-20)

Figure 2. Network structure (SARShipNet-20)

下载: 全尺寸图片幻灯片

图 3 卷积层内部操作流程

Figure 3. Internal operation flow in convolution layers

下载: 全尺寸图片幻灯片

图 4 通道注意力机制

Figure 4. Channel Attention (CA)

下载: 全尺寸图片幻灯片

图 5 空间注意力机制

Figure 5. Spatial Attention (SA)

下载: 全尺寸图片幻灯片

图 6 SARShipNet-20的SAR舰船检测结果

Figure 6. SAR ship detection results of SARShipNet-20

下载: 全尺寸图片幻灯片

图 7 SARShipNet-20性能评价曲线

Figure 7. Performance evaluation curve of SARShipNet-20

下载: 全尺寸图片幻灯片

表 1 SARShipNet-20的SAR舰船检测结果评价指标

Table 1. Evaluation index of SAR ship detection results of SARShipNet-20

类型	GT	TP	FN	FP	P_d (%)	P_m (%)	P_f (%)	Recall (%)	Precision (%)	mAP (%)	Time (ms)
T-CNN	184	180	4	8	97.83	2.17	4.26	97.83	95.74	96.88	10.14
DS-CNN	184	175	9	23	95.11	4.89	11.62	95.11	88.38	93.64	4.54
DS-CNN + CA	184	179	5	29	97.28	2.72	13.94	97.28	89.06	95.78	5.68
DS-CNN + SA	184	178	6	11	96.74	3.26	5.82	96.74	94.18	95.64	6.67
DS-CNN + CA + SA	184	180	4	8	97.83	2.17	4.26	97.83	95.74	96.93	8.72

下载: 导出CSV

表 2 不同方法的检测性能对比

Table 2. Comparison of detection performance of different methods

方法	P_d (%)	P_m (%)	P_f (%)	Recall (%)	Precision (%)	mAP (%)	Time (ms)
Faster R-CNN^[16]	85.16	14.84	18.85	85.16	81.15	82.66	327.48
RetinaNet^[34]	96.70	3.30	6.88	96.70	93.12	95.68	314.43
R-FCN^[35]	95.65	4.35	7.37	95.65	92.63	95.15	178.16
SSD^[18]	94.51	5.49	14.85	94.51	85.15	92.67	48.86
YOLOv3^[20]	96.70	3.30	6.38	96.70	93.62	95.34	22.30
YOLOv1^[28]	84.07	15.93	15.47	84.07	84.53	81.24	21.95
YOLOv2^[29]	92.86	7.14	15.08	92.86	84.92	90.09	19.01
YOLOv3-tiny^[20]	70.33	29.12	22.29	70.33	77.58	64.64	10.25
YOLOv2-tiny^[29]	47.80	52.20	26.27	47.80	73.73	44.40	9.43
SARShipNet-20(本文方法)	97.83	2.17	4.26	97.83	95.74	96.93	8.72

下载: 导出CSV

表 3 不同方法的模型对比

Table 3. Model comparison of different methods

方法	网络参数的数量	浮点运算量(FLOPs)	模型大小 (MB)
Faster R-CNN	272,746,867	545,429,460	752.75
RetinaNet	61,576,342	307,592,895	235.44
R-FCN	50,578,686	101,385,166	193.04
SSD	47,663,806	95,040,404	181.24
YOLOv3	36,382,957	72,545,184	139.25
YOLOv1	28,342,195	46,981,897,900	108.54
YOLOv2	23,745,908	118,685,133	90.73
YOLOv3-tiny	15,770,510	31,608,360	60.22
YOLOv2-tiny	8,676,244	86,692,284	33.20
SARShipNet-20(本文方法)	5,867,737	11,699,792	23.17

下载: 导出CSV

参考文献(38)

[1]	ZHANG Tianwen and ZHANG Xiaoling. High-speed ship detection in SAR images based on a grid convolutional neural network[J]. Remote Sensing, 2019, 11(10): 1206. doi: 10.3390/rs11101206
[2]	ZHANG Tianwen, ZHANG Xiaoling, SHI Jun, et al. Depthwise separable convolution neural network for high-speed SAR ship detection[J]. Remote Sensing, 2019, 11(21): 2483. doi: 10.3390/rs11212483
[3]	GAO Gui. A parzen-window-kernel-based CFAR algorithm for ship detection in SAR images[J]. IEEE Geoscience and Remote Sensing Letters, 2011, 8(3): 557–561. doi: 10.1109/LGRS.2010.2090492
[4]	AN Wentao, XIE Chunhua, and YUAN Xinzhe. An improved iterative censoring scheme for CFAR ship detection with SAR imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014, 52(8): 4585–4595. doi: 10.1109/TGRS.2013.2282820
[5]	HOU Biao, CHEN Xingzhong, and JIAO Licheng. Multilayer CFAR detection of ship targets in very high-resolution SAR images[J]. IEEE Geoscience and Remote Sensing Letters, 2015, 12(4): 811–815. doi: 10.1109/LGRS.2014.2362955
[6]	YIN Kuiying, JIN Lin, ZHANG Changchun, et al. A method for automatic target recognition using shadow contour of SAR image[J]. IETE Technical Review, 2013, 30(4): 313–323. doi: 10.4103/0256-4602.116721
[7]	JIANG Shaofeng, WANG Chao, ZHANG Bo, et al. Ship detection based on feature confidence for high resolution SAR images[C]. 2012 IEEE International Geoscience and Remote Sensing Symposium, Munich, Germany, 2012: 6844–6847. doi: 10.1109/IGARSS.2012.6352591.
[8]	WANG Shigang, WANG Min, YANG Shuyuan, et al. New hierarchical saliency filtering for fast ship detection in high-resolution SAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(1): 351–362. doi: 10.1109/TGRS.2016.2606481
[9]	WANG Chonglei, BI Funkun, CHEN Liang, et al. A novel threshold template algorithm for ship detection in high-resolution SAR images[C]. 2016 IEEE International Geoscience and Remote Sensing Symposium, Beijing, China, 2016: 100–103. doi: 10.1109/IGARSS.2016.7729016.
[10]	ZHU Jiwei, QIU Xiaolan, PAN Zongxu, et al. Projection shape template-based ship target recognition in TerraSAR-X images[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(2): 222–226. doi: 10.1109/LGRS.2016.2635699
[11]	LI Jianwei, QU Changwen, and SHAO Jiaqi. Ship detection in SAR images based on an improved faster R-CNN[C]. 2017 SAR in Big Data Era: Models, Methods and Applications, Beijing, China, 2017: 1–6. doi: 10.1109/BIGSARDATA.2017.8124934.
[12]	李健伟, 曲长文, 彭书娟. 基于级联CNN的SAR图像舰船目标检测算法[J]. 控制与决策, 2019, 34(10): 2191–2197. LI Jianwei, QU Changwen, and PENG Shujuan. A ship detection method based on cascade CNN in SAR images[J]. Control and Decision, 2019, 34(10): 2191–2197.
[13]	CHENG Mingming, LIU Yun, LIN Wenyan, et al. BING: Binarized normed gradients for objectness estimation at 300fps[J]. Computational Visual Media, 2019, 5(1): 3–20. doi: 10.1007/s41095-018-0120-1
[14]	SIMONYAN K and ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. arXiv: 1409.1556v1, 2014.
[15]	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. doi: 10.3969/j.issn.1001-506X.2018.09.09 LI Jianwei, QU Changwen, PENG Shujuan, et al. Ship detection in SAR images based on convolutional neural network[J]. Systems Engineering and Electronics, 2018, 40(9): 1953–1959. doi: 10.3969/j.issn.1001-506X.2018.09.09
[16]	REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137–1149. doi: 10.1109/TPAMI.2016.2577031
[17]	杨龙, 苏娟, 李响. 基于深度卷积神经网络的SAR舰船目标检测[J]. 系统工程与电子技术, 2019, 41(9): 1990–1997. doi: 10.3969/j.issn.1001-506X.2019.09.11 YANG Long, SU Juan, LI Xiang. Ship detection in SAR images based on deep convolutional neural network[J]. Systems Engineering and Electronics, 2019, 41(9): 1990–1997. doi: 10.3969/j.issn.1001-506X.2019.09.11
[18]	LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: Single shot multibox detector[C]. The 14th European Conference on Computer Vision, Amsterdam, Netherlands, 2016: 21–37. doi: 10.1007/978-3-319-46448-0_2.
[19]	胡昌华, 陈辰, 何川, 等. 基于深度卷积神经网络的SAR图像舰船小目标检测[J]. 中国惯性技术学报, 2019, 27(3): 397–405, 414. doi: 10.13695/j.cnki.12-1222/o3.2019.03.018 HU Changhua, CHEN Chen, HE Chuan, et al. Ship small target detection based on deep convolution neural network in SAR image[J]. Journal of Chinese Inertial Technology, 2019, 27(3): 397–405, 414. doi: 10.13695/j.cnki.12-1222/o3.2019.03.018
[20]	REDMON J and FARHADI A. YOLOv3: An incremental improvement[J]. arXiv: 1804.02767, 2018.
[21]	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 936–944. doi: 10.1109/CVPR.2017.106.
[22]	SIFRE L. Rigid-motion scattering for image classification[D]. [Ph.D. dissertation], Ecole Polytechnique, CMAP, 2014.
[23]	WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[C]. The 15th European Conference on Computer Vision, Munich, Germany, 2018: 3–19. DOI: 10.1007/978-3-030-01234-2_1.
[24]	HU Jie, SHEN Li, and SUN Gang. Squeeze-and-excitation networks[C]. The 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake, USA, 2018: 7132–7141. doi: 10.1109/CVPR.2018.00745.
[25]	HUBEL D H and WIESEL T N. Receptive fields of single neurones in the cat’s striate cortex[J]. The Journal of Physiology, 1959, 148(3): 574–591. doi: 10.1113/jphysiol.1959.sp006308
[26]	KRIZHEVSKY A, SUTSKEVER I, and HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84–90. doi: 10.1145/3065386
[27]	CHOLLET F. Xception: Deep learning with depthwise separable convolutions[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 1800–1807. doi: 10.1109/CVPR.2017.195.
[28]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 779–788. doi: 10.1109/CVPR.2016.91.
[29]	REDMON J and FARHADI A. YOLO9000: Better, faster, stronger[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 6517–6525. doi: 10.1109/CVPR.2017.690.
[30]	IOFFE S and SZEGEDY C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[C]. The 32nd International Conference on Machine Learning, Lille, French, 2015: 448–456.
[31]	MANASWI N K. Understanding and Working with Keras[M]. MANASWI N K. Deep Learning with Applications Using Python. Apress, Berkeley, CA: Springer, 2018: 31–43.
[32]	ABADI M, AGARWAL A, BARHAM P, et al. TensorFlow: Large-scale machine learning on heterogeneous systems. Software available from tensorflow.org[EB/OL]. https://www.bibsonomy.org/bibtex/2ba528cb1d5505ae48100cfc940c5fc3, 2015.
[33]	CHEN L C, PAPANDREOU G, SCHROFF F, et al. Rethinking atrous convolution for semantic image segmentation[J]. arXiv: 1706.05587, 2017.
[34]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]. 2017 IEEE International Conference on Computer Vision, Venice, Italy, 2017: 2999–3007. doi: 10.1109/ICCV.2017.324.
[35]	DAI Jifeng, HE Kaiming, and SUN Jian. R-FCN: Object detection via region-based fully convolutional networks[J]. arXiv: 1605.06409v2, 2016.
[36]	HE Kaiming, GIRSHICK R, and DOLLÁR P. Rethinking ImageNet pre-training[J]. arXiv: 1811.08883, 2018.
[37]	CUI Zongyong, LI Qi, CAO Zongjie, et al. Dense attention pyramid networks for multi-scale ship detection in SAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(11): 8983–8997. doi: 10.1109/TGRS.2019.2923988
[38]	孙显, 王智睿, 孙元睿, 等. AIR-SARShip-1.0: 高分辨率SAR舰船检测数据集[J]. 雷达学报, 2019, 8(6): 852–862. doi: 10.12000/JR19097 SUN Xian, WANG Zhirui, SUN Yuanrui, et al. AIR-SARShip-1.0: High resolution SAR ship detection dataset[J]. Journal of Radars, 2019, 8(6): 852–862. doi: 10.12000/JR19097

施引文献

期刊类型引用(26)

1.	吕奕龙，苟瑶，李敏，何玉杰，邢宇航. 基于注意力引导和多样本决策的舰船检测方法. 北京航空航天大学学报. 2025(01): 202-213 . 百度学术
2.	陆天宇，徐湛，崔红元，龚昊，王琤. 大幅宽SAR图像嵌入式舰船实时检测系统设计. 计算机工程与应用. 2024(01): 301-309 . 百度学术
3.	李生辉，李晓飞，宋璋晗，王必祥. 基于改进YOLOv5的船舶多尺度SAR图像检测算法. 数据采集与处理. 2024(01): 120-131 . 百度学术
4.	殷君君，罗嘉豪，李响，代晓康，杨健. 基于极化SAR梯度和复Wishart分类器的舰船检测. 雷达学报. 2024(02): 396-410 . 本站查看
5.	陈玮，刘坤. 基于卷积神经网络的SAR图像舰船分类. 计算机应用与软件. 2024(07): 159-164+183 . 百度学术
6.	李波，李志康，周钰彬. 结合特征融合和注意力机制的SAR舰船检测算法. 电子测量技术. 2024(10): 134-140 . 百度学术
7.	周文雪，张华春. 一种面向SAR图像快速舰船检测的轻量化网络. 中国科学院大学学报(中英文). 2024(06): 776-785 . 百度学术
8.	陈诗琪，王威，占荣辉，张军，刘盛启. 特征图知识蒸馏引导的轻量化任意方向SAR舰船目标检测器. 雷达学报. 2023(01): 140-153 . 本站查看
9.	张帆，陆圣涛，项德良，袁新哲. 一种改进的高分辨率SAR图像超像素CFAR舰船检测算法. 雷达学报. 2023(01): 120-139 . 本站查看
10.	邢世其，全斯农，范晖，王威，黄大通，李永祯，王雪松. 联合数学规划策略和精细极化分解的极化SAR舰船目标检测. 中国科学:信息科学. 2023(03): 585-605 . 百度学术
11.	甘文祥，张远谊，李欣园. 一种航空影像建筑物检测的轻量化CNN建模方法. 地理空间信息. 2023(06): 24-27 . 百度学术
12.	罗嘉豪，殷君君，杨健. 基于超像素与稀疏重构显著性的极化SAR舰船检测. 工程科学学报. 2023(10): 1684-1692 . 百度学术
13.	石晓荣，张康，倪亮，刘泽文，姜丰，陈鑫. 复杂环境下少样本域自适应雷达RD智能识别技术. 航天控制. 2023(04): 20-26 . 百度学术
14.	黄强，王钰宁，刘晓霞，胡云冰. 改进YOLOv3-SPP的SAR图像舰船目标检测. 遥感信息. 2023(05): 57-65 . 百度学术
15.	张天文，张晓玲，胥小我，邵子康，曾天娇. 基于显著性CNN的SAR图像靠岸舰船检测方法. 空天预警研究学报. 2023(04): 285-289 . 百度学术
16.	谭显东，彭辉. 改进YOLOv5的SAR图像舰船目标检测. 计算机工程与应用. 2022(04): 247-254 . 百度学术
17.	周玉金，谢宜壮，乔婷婷，冯杏. 基于Jetson TX2的SAR船只目标检测实现. 信号处理. 2022(02): 426-431 . 百度学术
18.	张云，化青龙，姜义成，徐丹. 基于混合型复数域卷积神经网络的三维转动舰船目标识别. 电子学报. 2022(05): 1042-1049 . 百度学术
19.	徐昌贵，张波，高建威，吴樊，张红，王超. FCOSR：一种无锚框的SAR图像任意朝向船舶目标检测网络. 雷达学报. 2022(03): 335-346 . 本站查看
20.	张晨，叶舟，吕宇宙，方明，高永婵. 基于ResNet的智能恒虚警目标检测方法. 上海航天(中英文). 2022(05): 71-78 . 百度学术
21.	刘涛，杨子渊，蒋燕妮，高贵. 极化SAR图像舰船目标检测研究综述. 雷达学报. 2021(01): 1-19 . 本站查看
22.	陈冬，句彦伟. 基于改进型YOLOv3的SAR图像舰船目标检测. 系统工程与电子技术. 2021(04): 937-943 . 百度学术
23.	毛强，晋良念，刘庆华. 穿墙雷达多维参数人体姿态识别方法. 雷达科学与技术. 2021(01): 40-47 . 百度学术
24.	周正，崔宗勇，曹宗杰，杨建宇. 基于特征转移金字塔网络的SAR图像跨尺度目标检测. 雷达学报. 2021(04): 544-558 . 本站查看
25.	郑慧敏. 基于矩阵分解的海洋SAR图像舰船检测. 计算机与现代化. 2020(09): 89-94 . 百度学术
26.	付晓雅，王兆成. 结合场景分类的近岸区域SAR舰船目标快速检测方法. 信号处理. 2020(12): 2123-2130 . 百度学术