«上一篇
 文章快速检索 高级检索

 哈尔滨工程大学学报  2019, Vol. 40 Issue (7): 1258-1262  DOI: 10.11990/jheu.201805057 0

### 引用本文

WANG Yanpeng, YANG Yang, YAO Yuan. Single shot multibox detector for ships detection in inland waterway[J]. Journal of Harbin Engineering University, 2019, 40(7), 1258-1262. DOI: 10.11990/jheu.201805057.

### 文章历史

Single shot multibox detector for ships detection in inland waterway
WANG Yanpeng , YANG Yang , YAO Yuan
School of Naval Architecture, Dalian University of Technology, Dalian 116024, China
Abstract: To avoid the excessive influence of external conditions on inland waterway environment by the traditional object detection algorithm, i.e., background modeling, a new method based on single shot multibox detector (SSD) is proposed for ship detection. The SSD model is based on the convolution neural network and uses the multi-scale regional features of the whole map to regress, so that the image can be used directly as the input of the network and false detection of external factors such as waves and leaf shaking can be avoided. Due to the shortage of ship samples in inland waterways, the method of data enhancement and migration learning was used to train the ship detection model, and this effectively alleviated the over-fitting phenomenon during the training process and yielded better detection results. A video detection of multiple groups of ships in different areas of the inland river shows that this method has better robustness and lower false detection rate than the traditional modeling algorithm. The ship recognition rates were over 90%, which was 16% higher than that of the traditional modeling algorithm.
Keywords: object detection    background modeling    inland waterway    convolutional neural network    single shot multibox detector (SSD)    data enhancement

1 SSD目标检测算法实现 1.1 SSD网络模型

1.2 损失函数

SSD网络模型的训练同时对位置和目标种类进行回归，其目标损失函数是置信度损失和位置损失之和，其表达式为：

 $L\left( {x, c, l, g} \right) = \frac{1}{N}\left( {{L_{{\rm{conf}}}}\left( {x, c} \right) + \alpha {L_{{\rm{loc}}}}\left( {x, l, g} \right)} \right)$ (1)

1.3 数据集构建

SSD把检测和分类一体化，实现端对端的训练。训练过程主要包括数据集构建和迁移学习应用。数据集包括训练集、验证集和测试集，采用了常州花园街大桥、淮安二堡船闸、无锡四河口、扬州茱萸湾等地的视频源。从视频源中提取1 000张图片作为训练集和验证集，后2个视频中的图片用作测试集。样本的采集要兼顾不同的时间段，保证样本的全面性，这对于最后检测器的泛化性能有重要影响。

1.4 迁移学习应用

2 实验结果分析

2.1 定性分析

2.2 定量评价指标对比

 $\left\{ \begin{array}{l} {P_r} = \frac{{{\rm{TP}}}}{{{\rm{TP}} + {\rm{FP}}}}\\ {R_e} = \frac{{{\rm{TP}}}}{{{\rm{TP}} + {\rm{TN}}}}\\ {F_1} = \frac{{2{\rm{TP}}}}{{{\rm{TP}} \times 2 + {\rm{FP + TN}}}} \end{array} \right.$ (2)

3 结论

1) 通过对训练的船舶样本做样本增强，可以丰富船舶图像训练集，更好的提取图像特征，泛化模型，也可以解决船舶样本数量不足的问题。

2) 利用迁移学习技术，先使用在Image Net竞赛中训练好模型的前几层参数来提取浅层特征，再加上标注好的船舶样本来训练新的模型。可以避免新的网络训练比较复杂，参数不好调整并且泛化能力不强的问题。

3) 相比于传统的目标检测方法，本文提出的SSD检测算法能够克服在波浪和岸边树叶晃动下造成的误检，其对不同清晰度、不同监控视角、不同船舶类型的场景具有很好的鲁棒性和实时性，工程适用性强。

 [1] 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 Jie, JIANG Zhiguo, ZHANG Haopeng, 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. (0) [2] BARRON J L, FLEET D J, BEAUCHEMIN S S. Performance of optical flow techniques[J]. International journal of computer vision, 1994, 12(1): 43-77. (0) [3] ALI A T, DAGLESS E L. Alternative practical methods for moving object detection[C]//Proceedings of 1992 International Conference on Image Processing and ITS Applications. Maastricht, 1992: 77-80. (0) [4] HARITAOGLU I, HARWOOD D, DAVIS L S. W4:real-time surveillance of people and their activities[J]. IEEE transactions on pattern analysis and machine intelligence, 2000, 22(8): 809-830. DOI:10.1109/34.868683 (0) [5] WREN C R, AZARBAYEJANI A, DARRELL T, et al. Pfinder: real-time tracking of the human body[C]//Proceedings of the Second International Conference on Automatic Face and Gesture Recognition. Killington, 1996: 780-785. https://www.media.mit.edu/publications/pfinder-real-time-tracking-of-the-human-body-2/ (0) [6] STAUFFER C, GRIMSON W E L. Adaptive background mixture models for real-time tracking[C]//Proceedings of 1999 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No PR00149). Fort Collins, 1999, 2: 246-252. https://wenku.baidu.com/view/511754bf1a37f111f1855b84.html (0) [7] BARNICH O, VAN DROOGENBROECK M. ViBe:a universal background subtraction algorithm for video sequences[J]. IEEE transactions on image processing, 2011, 20(6): 1709-1724. DOI:10.1109/TIP.2010.2101613 (0) [8] MEI Langqi, GUO Jianming, LU Pingping, et al. Inland ship detection based on dynamic group sparsity[C]//Proceedings of the 9th International Conference on Advanced Computational Intelligence. Doha, 2017: 1-6. (0) [9] REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[C]//Proceedings of the 28th International Conference on Neural Information Processing Systems. Montreal, 2015: 91-99. (0) [10] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, 2016: 779-788. (0) [11] LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[M]//LEIBE B, MATAS J, SEBE N, et al. Computer Vision-ECCV2016. Cham: Springer, 2016: 21-37. (0) [12] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[Z]. arXiv preprint arXiv: 1409.1556, 2014. (0)