基于改进SSD的水下光学图像感兴趣目标检测算法研究

李宝奇; 黄海宁; 刘纪元; 刘正君; 韦琳哲

doi:10.11999/JEIT210761

基于改进SSD的水下光学图像感兴趣目标检测算法研究

doi: 10.11999/JEIT210761

1.
中国科学院声学研究所北京 100190
2.
中国科学院先进水下信息技术重点实验室北京 100190

基金项目: 国家自然科学基金(11904386)，国家基础科研计划重大项目(JCKY2016206A003), 中国科学院青年创新促进会(2019023)

详细信息

作者简介:
李宝奇：男，特别研究助理，研究方向为水声信号处理、目标检测/识别和跟踪、深度学习理论

黄海宁：男，研究员，研究方向为水声信号与信息处理、目标探测、水声通信与网络等

刘纪元：男，研究员，研究方向为水声信号处理、数字信号处理和水声成像与图像处理等

刘正君：女，助理研究员，研究方向为水声信号处理等

韦琳哲：男，助理研究员，研究方向为水声信号处理等

通讯作者:
黄海宁　hhn@mail.ioa.ac.cn

中图分类号: TN911.73; TP391
计量
- 文章访问数: 951
- HTML全文浏览量: 593
- PDF下载量: 184
- 被引次数: 0
出版历程
- 收稿日期: 2021-07-30
- 修回日期: 2021-11-15
- 录用日期: 2021-11-18
- 网络出版日期: 2021-11-19
- 刊出日期: 2022-10-19

Underwater Optical Image Interested Object Detection Model Based on Improved SSD

1.
Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China
2.
Key Laboratory of Science and Technology on Advanced Underwater Acoustic Signal Processing, Chinese Academy of Sciences, Beijing 100190, China

Funds: The National Natural Science Foundation of China (11904386), The Major Projects of the National Basic Scientific Research Plan (JCKY2016206A003), The Youth Innovation Promotion Association of Chinese Academy of Sciences (2019023)

摘要

摘要: 针对轻量化目标模型SSD-MV2对水下光学图像感兴趣目标检测精度低的问题，该文提出一种通道可选择的轻量化特征提取模块(SEB)和一种卷积核可变形、通道可选择的特征提取模块(SDB)。与此同时，利用SEB模块和SDB模块分别重新设计了SSD-MV2的基础网络和附加特征提取网络，记作SSD-MV2SDB，并为其选择了合理的基础网络扩张系数和附加特征提取网络SDB模块数量。在水下图像感兴趣目标检测数据集UOI-DET上，SSD-MV2SDB比SSD-MV2检测精度提高3.04%。实验结果表明，SSD-MV2SDB适用于水下图像感兴趣目标检测任务。
- 水下光学图像感兴趣目标检测 /
- SSD /
- MobileNet V2 /
- 可变形卷积 /
- 通道可选择
Abstract: In order to solve the problem of low detection accuracy of SSD-MV2, a Selective and Efficient Block (SEB) and a Selective and Deformable Block (SDB) are proposed. At the same time, the basic network and additional feature extraction network of SSD-MV2 are redesigned by using SEB and SDB, which is named SSD-MV2SDB, and a set of reasonable expansion coefficient of basic network and number of SDB in additional feature extraction network are selected for SSD-MV2SDB. On UOI-DET, mAP of SSD-MV2SDB is 3.04% higher than that of SSD-MV2. The experimental results show that SSD-MV2SDB is suitable for underwater optical image interested object detection task.
- Underwater optical image interested object detection /
- Single Shot Detection (SSD) /
- MobileNet V2 /
- Deformable convolution /
- Channel selectable

HTML全文

图 1 SEB模块和SDB模块

下载: 全尺寸图片幻灯片

图 2 SSD-MV2SDB目标检测模型

下载: 全尺寸图片幻灯片

图 3 SSD-MV2SDB对水下光学图像感兴趣目标的检测效果图

下载: 全尺寸图片幻灯片

表 1 水下图像目标检测数据集组成(幅)

目标	训练	测试
方框	203	19
渔网	221	13
蛙人	214	26
UUV	194	22
球体	203	20
总计	1035	100

下载: 导出CSV

表 2 目标检测模型性能比较

模型	基础网络	附加特征提取网络	检测精度(%)	参数大小(MB)	检测时间(ms)
SSD-MV2	IRB	IRB	94.24	10.2	7.20
SSD-MV2SEB	SEB	SEB	95.09	11.0	10.01
SSD-MV2IRBD	SEB	IRBD	95.97	14.8	13.52
SSD-MV2SDB	SEB	SDB	97.28	14.9	13.86

下载: 导出CSV

表 3 基础网络扩张系数对SSD-MV2SDB性能的影响

扩张系数	检测精度(%)	参数大小(MB)	检测时间(ms)
2	95.03	12.1	13.66
4	97.28	14.9	13.86
6	97.33	17.7	13.90
8	97.76	20.4	14.12

下载: 导出CSV

表 4 附加特征提取网络SDB模块数量对SSD-MV2SDB性能的影响

模块数量	检测精度(%)	参数大小(MB)	检测时间(ms)
0	95.09	11.0	10.01
1	96.08	13.5	11.11
2	97.09	14.2	12.53
3	97.28	14.9	13.86

下载: 导出CSV

参考文献(19)

[1]	YEH C H, LIN Chuhan, KANG Liwei, et al. Lightweight deep neural network for joint learning of underwater object detection and color conversion[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 6: 1–15. doi: 10.1109/TNNLS.2021.3072414
[2]	HMUE P M and PUMRIN S. Image enhancement and quality assessment methods in turbid water: A review article[C]. IEEE International Conference on Consumer Electronics, Bangkok, Thailand, 2019: 59–63.
[3]	CHEN Bin, LI Rong, BAI Wanjian, et al. Research on recognition method of optical detection image of underwater robot for submarine cable[C]. 2019 IEEE 3rd Advanced Information Management, Communicates, Electronic and Automation Control Conference (IMCEC), Chongqing, China, 2019: 1973–1976.
[4]	LECUN Y, BENGIO Y, and HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436–444. doi: 10.1038/nature14539
[5]	ZHANG J X, YORDANOV B, GAUNT A, et al. A deep learning model for predicting next-generation sequencing depth from DNA sequence[J]. Nature Communications, 2021, 12: 4387. doi: 10.1038/s41467-021-24497-8
[6]	WANG Shiqiang. Efficient deep learning[J]. Nature Computational Science, 2021, 1(3): 181–182. doi: 10.1038/s43588-021-00042-x
[7]	LAGEMANN C, LAGEMANN K, MUKHERJEE S, et al. Deep recurrent optical flow learning for particle image velocimetry data[J]. Nature Machine Intelligence, 2021, 3(7): 641–651. doi: 10.1038/s42256-021-00369-0
[8]	LI Sichun, JIN Xin, YAO Sibing, et al. Underwater small target recognition based on convolutional neural network[C]. Global Oceans 2020: Singapore – U. S. Gulf Coast, Biloxi, USA, 2020: 1–7.
[9]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]. The IEEE Conference on Computer Vision and Pattern Recognition, Columbus, USA, 2014: 580–587.
[10]	GIRSHICK R. Fast R-CNN[C]. IEEE International Conference on Computer Vision (ICCV), Santiago, Chile, 2015: 1440–1448.
[11]	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
[12]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, USA, 2016: 779–788.
[13]	SRITHAR S, PRIYADHARSINI M, SHARMILA F M, et al. Yolov3 Supervised machine learning framework for real-time object detection and localization[J]. Journal of Physics:Conference Series, 2021, 1916: 012032. doi: 10.1088/1742-6596/1916/1/012032
[14]	IANDOLA F N, MOSKEWICZ N W, ASHRAF K, et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1 MB model size[EB/OL]. https://arxiv.org/abs/1602.07360v1, 2016.
[15]	SZEGEDY C, LIU Wei, JIA Yangqing, et al. . Going deeper with convolutions[C]. IEEE Conference on Computer Vision and Pattern Recognition, Boston, USA, 2015: 1–9.
[16]	HOWARD A G, ZHU Menglong, CHEN Bo, et al. MobileNets: Efficient convolutional neural networks for mobile vision applications[EB/OL]. https://arxiv.org/abs/1704.04861, 2017.
[17]	SANDLER M, HOWARD A, ZHU Menglong, et al. MobileNetV2: Inverted residuals and linear bottlenecks[C]. The IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 4510–4520.
[18]	HU Jie, SHEN Li, ALBANIE S, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011–2023. doi: 10.1109/TPAMI.2019.2913372
[19]	DAI Jifeng, QI Haozhi, XIONG Yuwen, et al. Deformable convolutional networks[C]. IEEE International Conference on Computer Vision, Venice, Italy, 2017: 764–773.