﻿ 船舶移动目标高效识别算法
 舰船科学技术  2022, Vol. 44 Issue (24): 145-148    DOI: 10.3404/j.issn.1672-7649.2022.24.030 PDF

Efficient recognition algorithm for ship moving target
LIU Cui-huan
Software College of Hebei Polytechnic Institute, Shijiazhuang 050090, China
Abstract: In recent years, the rise of unmanned ships has changed the traditional manual operation mode, unmanned driving, digital driving, intelligent driving will become the main trend of the future ship development. In order to ensure the safety of unmanned ships, ships should have the ability to accurately perceive the surrounding environment, and use a variety of sensors and artificial intelligence technologies to establish a Marine mobile target detection system to ensure their safe operation at sea. Multi-feature fusion technology for detection has a much higher accuracy than the general neural network, and has a great improvement compared with the traditional neural network methods. Finally, multiple moving objects in the ocean are identified, and the results show that the proposed algorithm is a stable and effective method.Therefore, this algorithm is worth popularizing in the future.
Key words: moving target     identification algorithm     ship
0 引　言

1 海上船舶移动目标分类与数据分析 1.1 目标识别常用数据集

 图 1 海上船舶移动目标 Fig. 1 Marine ship movement target

 图 2 海上船舶静态目标 Fig. 2 Static target of ships at sea
1.2 海上目标数据增广

 $\phi ：\delta \to \tau \text{，}$

 ${\delta ^ \bullet } = \delta \cup \tau 。$

 $\left[ {\begin{array}{*{20}{c}} {{{x}^ \bullet }} \\ {{{y}^ \bullet }} \\ 1 \end{array}} \right] = T\left[ {\begin{array}{*{20}{c}} {{{x}^{}}} \\ {{{y}^{}}} \\ 1 \end{array}} \right] = \left[ {\begin{array}{*{20}{c}} {{\alpha _{11}}}&{{\alpha _{12}}}&0 \\ {{\alpha _{21}}}&{{\alpha _{22}}}&0 \\ {{\alpha _{31}}}&{{\alpha _{32}}}&1 \end{array}} \right]\left[ {\begin{array}{*{20}{c}} {x} \\ {y} \\ 1 \end{array}} \right] 。$

 $\left[ {\begin{array}{*{20}{c}} {{{x}^ \bullet }} \\ {{{y}^ \bullet }} \\ 1 \end{array}} \right] = \left[ {\begin{array}{*{20}{c}} 1&0&{\Delta {x}} \\ 0&1&{\Delta {y}} \\ 0&0&1 \end{array}} \right]\left[ {\begin{array}{*{20}{c}} {x} \\ {y} \\ 1 \end{array}} \right] \text{，}$

 $\left[ {\begin{array}{*{20}{c}} {{{x}^ \bullet }} \\ {{{y}^ \bullet }} \\ 1 \end{array}} \right] = \left[ {\begin{array}{*{20}{c}} {{\text{cos}}\theta }&{ - {\text{sin}}\theta }&0 \\ {{\text{sin}}\theta }&{{\text{cos}}\theta }&0 \\ 0&0&1 \end{array}} \right]\left[ {\begin{array}{*{20}{c}} {x} \\ {y} \\ 1 \end{array}} \right] 。$

 图 3 左面镜像翻转 Fig. 3 Left-side mirror flip

 图 4 右面镜像翻转 Fig. 4 Right-face mirror flip
2 海上船舶移动目标识别算法 2.1 卷积神经网络

2.2 AlexNet

AlexNet的网络结构主要包括3个完全连接层和5个卷积层，在部分卷积层的后方设有一个池化层，采用Max法，得出图5的网络结构图。

 图 5 AlexNet结构图 Fig. 5 AlexNet structure diagram

2.3 海上目标识别相关算法

 ${\theta }_{{t}+1}={\theta }_{{t}}-\eta \cdot {\nabla }_{\theta }J(\theta ,{{x}}^{({ii}+{n})},{{y}}^{({ii}+{n})})。$

3 特征融合海上船舶移动目标识别

3.1 多层次特征融合的网络结构

SSD的网络结构类似于图像金字塔，把第4个卷积层和第5个卷积层的最后一层的输出特性相结合，即把Conv4_3和Conv5_3相结合。图6为VGGNet-16融合后的网络结构。

 图 6 多层次网络结构图 Fig. 6 Multi-level network structure diagram

3.2 迁移学习

1）构建的新网络更稳定。将训练有素的网络模型的一些参数添加到训练中，将比从零开始构建和训练一个新网络模型更加稳定和强大；

2）网络培训成本更低。使用一些迁移的网络参数进行培训可以节省大量的人力、物力和时间成本；

3）适用于小样本训练。由于网络越大，对数据的需求就越大，因此该方法可以避免过度适应训练。

4 结　语

 [1] 王梁, 侯义锋, 贺杰. 基于Mask-RCNN的自然场景下油茶果目标识别与检测[J]. 中国农机化学报, 2022, 43(12): 148-15+189. [2] 康志强, 张思乾, 封斯嘉, 等. 稀疏先验引导CNN学习的SAR图像目标识别方法[J/OL]. 信号处理, 2022: 1–15. KANG Zhi-qiang, ZHANG Si-qian, FENG Si-jia, et al. SAR Image Target Recognition Method Based on Sparse Prior Guided CNN Learning[J/OL]. Signal Processing: 1–15. [3] 陈瑾龙, 徐哲壮, 黄平, 等. 基于机器视觉的四足机器人目标识别与跟随系统设计[J]. 实验技术与管理, 2022, 39(10): 135–139+167. [4] 王宏伟, 杨力, 张方雨. 基于神经网络的电磁感应探雷目标识别方法[J]. 探测与控制学报, 2022, 44(5): 90-96. WANG Hong-wei, YANG Li, ZHANG Fang-yu. Target recognition method for electromagnetic induction mine detection based on neural network[J]. Journal of Detection and Control, 2022, 44(5): 90-96. [5] 李建江, 陈玮, 李明, 等. 基于网格热度值的船舶规律路径提取算法[J]. 计算机研究与发展, 2018, 055(005): 908-919. DOI:10.7544/issn1000-1239.2018.20170226 [6] 朱岱寅, 耿哲, 俞翔, 等. 地面目标多角度SAR数据集构建与目标识别方法[J]. 南京航空航天大学学报, 2022, 54(5): 985-994. ZHU Dai-yin, GENG Zhe, YU Xiang, HAN Sheng-liang, YANG Wei-xing, LV Ji-ming, YE Zheng, YAN He. Ground target multi-angle SAR data set construction and target recognition method[J]. Journal of Nanjing University of Aeronautics and Astronautics, 2022, 54(5): 985-994.