基于深浅层特征融合的舰船要害关键点检测算法

doi:10.12305/j.issn.1001-506X.2021.11.24

摘要/Abstract

摘要：

导引头对舰船要害部位的精确检测能力是精确制导武器的核心技术之一。针对导引头对舰船要害部位检测精度低、模型参数冗余、相对运动导致舰船图像尺度、角度变化剧烈等问题, 提出了基于深浅层特征融合的舰船要害关键点检测算法。首先，采用多尺度特征融合模块融合不同感受野的有效信息; 其次，利用SoftPool池化改善下采样导致的信息损失, 利于区分相似关键点; 然后，引入深度可分离卷积降低参数冗余, 结合轻量化注意力机制增强有效特征表达; 最后，利用在线难例挖掘改善样本不均衡, 提升收敛效果。改进后的舰船要害关键点检测算法准确率提升4.4%, 算法兼具检测精度与速度优势, 鲁棒性较好。

关键词: 关键点检测, 特征融合, 反舰导弹

Abstract:

The precision detection capability of seeker to target the warship's vital parts is one of the core technologies of precision-guidance weapons. Aiming at the problems of seeker's low detection accuracy, redundant model parameters and drastic changes of images' scale and angle caused by relative motion, the key-points detection algorithm based on fusion of deep and shallow features for warship's vital part is proposed. Firstly, the multi-scale features fusion module is used to fuse the effective information of different receptive fields. Secondy, SoftPool is used to reduce the information loss caused by down sampling, which is conducive to distinguish similar key-points. Then, the depthwise separable convolution is introduced to decrease the parameter redundancy, and the lightweight attention mechanism is combined to enhance the expression of effective features. Finally, the online hard example mining is applied to improve the sample imbalance and accelerate the convergence. The accuracy rate of improved warship's key-points detection algorithm increases by 4.4%. The algorithm not only has the advantages of detection accuracy and detection speed, but also has better robustness.

Key words: key-points detection, feature fusion, anti-ship missiles

中图分类号:

李晨瑄, 钱坤, 胥辉旗. 基于深浅层特征融合的舰船要害关键点检测算法[J]. 系统工程与电子技术, 2021, 43(11): 3239-3249.

Chenxuan LI, Kun QIAN, Huiqi XU. Key-points detection algorithm based on fusion of deep and shallow features for warship's vital part[J]. Systems Engineering and Electronics, 2021, 43(11): 3239-3249.

图/表 17

图1

图2

图3

图4

图5

图6

图7

图8

表1

表2

图9

图10

表3

表4

图11

表5

图12

参考文献 29

1	余瑞星, 吴虞霖, 曹萌, 等. 基于边缘与角点相结合的目标提取与匹配算法[J]. 西北工业大学学报, 2017, 35 (4): 586- 590. doi: 10.3969/j.issn.1000-2758.2017.04.005
	YU R X , WU Y L , CAO M , et al. Target extraction and image matching algorithm based on combination of edge and corner[J]. Journal of Northwestern Polytechnical University, 2017, 35 (4): 586- 590. doi: 10.3969/j.issn.1000-2758.2017.04.005
2	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.
3	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.
4	张迪飞, 张金锁, 姚克明, 等. 基于SVM分类的红外舰船目标识别[J]. 红外与激光工程, 2016, 45 (1): 179- 184.
	ZHANG D F , ZHANG J S , YAO K M , et al. Infrared ship-target recognition based on SVM classification[J]. Inrared and Laser Engineering, 2016, 45 (1): 179- 184.
5	KANG M, LENG X G, LIN Z, et al. A modified faster R-CNN based on CFAR algorithm for SAR ship detection[C]//Proc. of the IEEE Conference on International Workshop on Remote Sensing with Intelligent Processing, 2017.
6	WANG Y Y , WANG C , ZHANG H , et al. Automatic ship detection based on RetinaNet using multi-resolution Gaofen-3 ima-gery[J]. Remote Sensing, 2019, 11 (5): 531- 545. doi: 10.3390/rs11050531
7	LIN T , GOYAL P , GIRSHICK R , et al. Focal loss for dense object detection[J]. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2020, 42 (2): 318- 327. doi: 10.1109/TPAMI.2018.2858826
8	李晖晖, 周康鹏, 韩太初. 基于CReLU和FPN改进的SSD舰船目标检测[J]. 仪器仪表学报, 2020, 41 (4): 183- 190.
	LI H H , ZHOU K P , HAN T C . Ship object detection based on SSD improved with CReLU and FPN[J]. Chinese Journal of Scientific Instrument, 2020, 41 (4): 183- 190.
9	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multiBox detector[C]//Proc. of the Conference on European Conference on Computer Vision, 2016: 21-37.
10	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.
	YANG L , SU J , LI X . Ship detection in SAR images based on deep convolutional neural network[J]. Systems Engineering and Electronics, 2019, 41 (9): 1990- 1997.
11	CHEN Y C , TIAN Y L , HE M Y . Monocular human pose estimation: a survey of deep learning-based methods[J]. Computer Vision and Image Understanding, 2020, 192, 224- 244.
12	PAPANDREOU G, ZHU T, KANAZAWA N, et al. Towards accurate multi-person pose estimation in the wild[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 3711-3719.
13	HE K, GKIOXARI G, DOLLAR P, et al. Mask R-CNN[C]//Proc. of the IEEE International Conference on Computer Vision, 2017: 2980-2988.
14	XIAO B, WU H P, WEI Y C, et al. Simple baselines for human pose estimation and tracking[C]//Proc. of the IEEE conference on European Conference on Computer Vision, 2018: 472-487.
15	CAO Z, SIMON T, WEI S, et al. Realtime multi-person 2D pose estimation using part affinity fields[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 1302-1310.
16	PISHCHULIN L, INSAFUTDINOV E, TANG S, et al. DeepCut: joint subset partition and labeling for multi person pose estimation[C]//Proc. of the Conference on Computer Vision and Pattern Recognition, 2016: 4229-4937.
17	NEWELL A, HUANG Z A, DENG J, et al. Associative embedding: end-to-end learning for joint detection and grouping[C]//Proc. of the Neural Information Processing Systems, 2017: 2277-2287.
18	XU J J , SONG B , YANG X , et al. An improved deep keypoint detection network for space targets pose estimation[J]. Remote Sensing, 2020, 12, 3857- 3878. doi: 10.3390/rs12233857
19	ZHOU X Y, WANG D Q, PHILIPP K, et al. Objects as points[EB/OL]. [2021-02-02]. https://arxiv.org/abs/1904.07850.
20	XIAO B, WU H P, WEI Y C. Simple baselines for human pose estimation and tracking[C]//Proc. of the IEEE Conference on European Conference on Computer Vision, 2018: 472-487.
21	YU F, WANG D Q, SHELHAMER E, et al. Deep layer aggregation[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 2403-2412.
22	NEWELL A, YANG K Y, DENG J. Stacked hourglass networks for human pose estimation[EB/OL][2021-02-02]. https://arxiv.org/abs/1603.06937.
23	ALEXANDROS S, RONALD P, GRIGORIOS K. Refining activation downsampling with SoftPool[EB/OL]. [2021-02-02]. https://arxiv.org/abs/2101.00440.
24	BRIJRAJ S , DURGA T , KUMAR A S . Shunt connection: an intelligent skipping of contiguous blocks for optimizing MobileNet-V2[J]. Neural Networks, 2019, 118, 192- 203. doi: 10.1016/j.neunet.2019.06.006
25	WANG Q L, WU B G, ZHU P F, et al. Efficient channel attention for deep convolutional neural networks[C]//Proc. of the Conference on Computer Vision and Pattern Recognition, 2019.
26	SHRIVASTAVA A, GUPTA A, GIRSHICK R. Training region-based object detectors with online hard example mining[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 761-769.
27	ZHOU Y Q , CAI Z M , ZHU Y T , et al. Automatic ship detection in SAR image based on multi-scale faster R-CNN[J]. Journal of Physics: Conference Series, 2020, 1550 (4): 042006. doi: 10.1088/1742-6596/1550/4/042006/pdf
28	黄洁, 姜志国, 张浩鹏, 等. 基于卷积神经网络的遥感图像舰船目标检测[J]. 北京航空航天大学学报, 2017, 43 (9): 1841- 1848.
	HUANG J , JIANG Z G , ZHANG H P , et al. Ship object detection in remote sensing images using convolutional neural networks[J]. Journal of Beijing University of Aeronautics and Astronautics, 2017, 43 (9): 1841- 1848.
29	张雪松, 庄严, 闫飞, 等. 基于迁移学习的类别级物体识别与检测研究与进展[J]. 自动化学报, 2019, 45 (7): 1224- 1243.
	ZHANG X S , ZHUANG Y , YAN F , et al. Status and devel-opment of transfer learning based category-level object recognition and detection[J]. Acta Automatica Sinica, 2019, 45 (7): 1224- 1243.

参数	配置信息
CPU	AMD Ryzen 9 3900X
CPU显存	32G
GPU	GEFORCE RTX 2080Ti
GPU显存	11G
IDE	Pycharm、gedit、vim
系统	Ubuntu 16.04 LTS
语言	Python
加速环境	CUDA10.0, CuDNN7.6
深度学习框架	Pytorch1.0

算法	mAP/%	召回率/%	FPS
Resnet18	81.3	82.5	25
Res-dcn18	80.8	81.9	42
DLA34	82.8	84.1	37
DLA-dcn	83.2	84.9	26
Hourglass	86.0	87.1	12
所提方法	87.6	88.5	27

算法	参数/M	模型规模/M
Resnet18	15.82	65.11
Res-dcn18	14.43	58.26
DLA34	18.16	75.76
DLA-dcn	20.17	80.64
Hourglass	191.25	779.88
本文方法	19.51	73.4

DLA-dcn	Liblock	SoftPool	RFB+ECA	参数/M	模型规模/M
√	—	—	—	20.17	80.64
√	√	—	—	17.84	64.8
√	—	√	—	20.17	87.2
√	—	—	√	20.19	82.6
√	√	√	—	18.73	70.1
√	√	—	√	19.84	72.9
√	—	√	√	20.19	82.8
√	√	√	√	19.51	73.4

DLA-dcn	Liblock	SoftPool	RFB+ECA	mAP/%	召回率/%	FPS
√	—	—	—	83.2	84.9	26
√	√	—	—	82.9	84.0	31
√	—	√	—	85.0	86.2	28
√	—	—	√	86.8	88.0	28
√	√	√	—	87.1	88.3	30
√	√	—	√	87.0	88.1	28
√	—	√	√	87.5	88.5	29
√	√	√	√	87.6	88.5	27