Ship Detection in SAR Images Based on Knowledge Distillation

doi:10.3969/j.issn.1009-086x.2023.04.010

Abstract

Abstract:

With the rapid development of deep learning， the object detection network model hsa achieved great success in ship detection from synthetic aperture radar （SAR） images. In order to achieve better results， a large detection network is usually adopted， which requires more computing resources and slower inference. Knowledge distillation can effectively compress the network， but most of them are for image classification. Considering the difference between ships and background in SAR images， this paper proposes a combining local and global distillation method for ship detection. Considering the scattering characteristics of ships in SAR images， slice preprocessing in the amplitude direction is performed on the ship sample data constructing a separate data channel with more significant scattering characteristics， which improves the quality of input data fed to the network. Experimental results based on SAR ship detection dataset （SSDD） show that the proposed approach can effectively reduce the price of the network model and improve the detection performance， achieving 90.7% mAP.

Key words: synthetic aperture radar（SAR）, object detection, ship, deep learning, knowledge distillation

摘要：

随着深度学习的快速发展，目标检测网络模型在SAR图像舰船检测取得了很大的成功。为了取得更好的效果，往往采取规模很大的检测网络，这需要更多的计算资源并且推理更慢。知识蒸馏能够有效地压缩网络，但当前的知识蒸馏大都是对于图像分类问题。在分析SAR图像中舰船与背景差异的基础上，提出局部与全局蒸馏的舰船目标检测方法。考虑到舰船在SAR图像中的散射特性，对舰船样本数据进行幅值方向的切片预处理，作为具有更显著散射特性的单独数据通道，提升了网络输入数据的质量。利用SSDD(SAR ship detection dataset)舰船数据集开展目标检测实验和评估。实验结果表明，所提出的方法能够在轻量化网络下，提升舰船检测性能，mAP能达到90.7%。

关键词: 合成孔径雷达, 目标检测, 舰船, 深度学习, 知识蒸馏

CLC Number:

TN957.52

Yang LI, Xiao LIANG, Haisa JIA. Ship Detection in SAR Images Based on Knowledge Distillation[J]. Modern Defense Technology, 2023, 51(4): 78-85.

李洋, 梁潇, 贾海萨. 基于知识蒸馏的SAR图像舰船目标检测[J]. 现代防御技术, 2023, 51(4): 78-85.

Figures/Tables 9

References 18

1	WANG Chonglei， BI Fukun， ZHANG Weiping， et al. An Intensity-Space Domain CFAR Method for Ship Detection in HR SAR Images ［J］. IEEE Geoscience and Remote Sensing Letters， 2017， 14（4）： 529-533.
2	LENG Xiangguang， JI Kefeng， YANG Kai， et al. A Bilateral CFAR Algorithm for Ship Detection in SAR Images［J］. IEEE Geoscience and Remote Sensing Letters， 2015， 12（7）： 1536-1540.
3	PAPPAS O， ACHIM A， BULL D. Superpixel-Level CFAR Detectors for Ship Detection in SAR Imagery［J］. IEEE Geoscience and Remote Sensing Letters， 2018， 15（9）： 1397-1401.
4	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.
5	CAI Zhaowei， VASCONCELOS N. Cascade R-Cnn： Delving into High Quality Object Detection［C］∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018： 6154-6162.
6	LIN Sungyi， GOYAL P， GIRSHICK R， et al. Focal Loss for Dense Object Detection［C］∥Proceedings of the IEEE International Conference on Computer Vision. 2017： 2980-2988
7	LIU Wei， ANGUELOV D， ERHAN D， et al. Ssd： Single Shot Multibox Detector［C］∥European Conference on Computer Vision. Springer， Cham， 2016： 21-37.
8	ZHOU Xingyi， WANG Dequan， Krähenbühl P. Objects as Points［J］. arXiv preprint arXiv：1904.07850， 2019：2103-2112.
9	TIAN Zhi， SHEN Chunhua， CHEN Hao， et al. Fcos： Fully Convolutional One-Stage Object Detection［C］∥Proceedings of the IEEE/CVF International Conference on Computer Vision， 2019： 9627-9636.
10	JIAO Jiao， ZHANG Yue， SUN Hao， et al. A Densely Connected End-to-End Neural Network for Multiscale and Multiscene SAR Ship Detection［J］. IEEE Access， 2018， 6： 20881-20892.
11	FU Jiamei， SUN Xian， WANG Zhirui， et al. An Anchor-Free Method Based on Feature Balancing and Refinement Network for Multiscale Ship Detection in SAR Images［J］. IEEE Transactions on Geoscience and Remote Sensing， 2020， 59（2）： 1331-1344.
12	CUI Zongyong， WANG Xiaoya， LIU Nengyuan， et al. Ship Detection in Large-Scale SAR Images Via Spatial Shuffle-Group Enhance Attention［J］. IEEE Transactions on Geoscience and Remote Sensing， 2020， 59（1）： 379-391.
13	HINTON G， VINYALS O， DEAN J. Distilling the Knowledge in a Neural Network［J］. arXiv Preprint arXiv：1503.02531， 2015， 2（7）：21-29.
14	CHEN Guobin， CHOI Wongun， YU Xiang， et al. Learning Efficient Object Detection Models with Knowledge Distillation［J］. Advances in Neural Information Processing Systems， 2017， 30（2）：742-751.
15	LI Quanquan， JIN Shengying， YAN Junjie. Mimicking Very Efficient Network for Object Detection［C］∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017： 6356-6364.
16	ZHANG Linfeng， MA Kaisheng. Improve Object Detection with Feature-Based Knowledge Distillation： Towards Accurate and Efficient Detectors［C］∥International Conference on Learning Representations， 2020：3327-3336.
17	YANG Zhendong， LI Zhe， JIANG Xiaohu， et al. Focal and Global Knowledge Distillation for Detectors［C］∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2022： 4643-4652.
18	CAO Yue， XU Jiarui， LIN Stephen， et al. Gcnet： Non-Local Networks Meet Squeeze-Excitation Networks and Beyond［C］∥Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops， 2019：1971-1980.

方法	切片	蒸馏	P	R	F1	mAP
基线模型	-	-	0.871	0.944	0.906	0.891
	√	-	0.906	0.923	0.912	0.897
	-	√	0.887	0.938	0.912	0.901
本文模型	√	√	0.921	0.912	0.916	0.907

方法	切片	蒸馏	P	R	F1	mAP
基线模型	-	-	0.871	0.944	0.906	0.891
	√	-	0.906	0.923	0.912	0.897
	-	√	0.887	0.938	0.912	0.901
本文模型	√	√	0.921	0.912	0.916	0.907

网络模型	骨干网络	P	R	F1	mAP
本文模型	ResNet50	0.887	0.938	0.912	0.901
教师	ResNet101	0.943	0.968	0.955	0.903
未蒸馏学生	ResNet50	0.871	0.944	0.906	0.891

网络模型	骨干网络	P	R	F1	mAP
本文模型	ResNet50	0.887	0.938	0.912	0.901
教师	ResNet101	0.943	0.968	0.955	0.903
未蒸馏学生	ResNet50	0.871	0.944	0.906	0.891

蒸馏	局部	全局	P	R	F1	mAP
基线模型	-	-	0.871	0.944	0.907	0.891
	√	-	0.906	0.923	0.914	0.898
	-	√	0.902	0.926	0.913	0.896
本文模型	√	√	0.887	0.938	0.912	0.901