﻿ 基于RCS统计特征的船舶目标识别方法
 舰船科学技术  2018, Vol. 40 Issue (7): 129-132 PDF

The target identification method for ship based on RCS statistical feature
JI Yong-qiang, LIU Tong, XU Gao-zheng, SHI Yu-hao, ZHANG Yu-ping, YANG Jin-hong
Systems Engineering Research Institute of CSSC, Beijing 100190, China
Abstract: The Radar Cross Section (RCS) is an important parameter for measuring the ship target scatter characteristics and the most effective electromagnetic spectrum characteristics for the classification and recognition of the ship.The shape and structure of ship are complex and the mechanism of the electromagnetic scattering is complex. At the same time, it is influenced by the detection angle of radar and the electromagnetic environment in the sea area. So the RCS of ship shows obvious fluctuation characteristics. In this paper, the statistical characteristics of ships’ RCS is conducted, and the BP neural network is used for the recognition of ships.The results show that the method proposed in the paper achieved good results and it can accurately identify different types of ship under certain condtions.
Key words: RCS     statistical characteristics     BP neural network     ship target identification
0 引　言

1 船舶RCS统计特征分析

1.1 位置特征参数（均值、方差和极差）

 $\bar \sigma = \frac{1}{n}\sum\limits_{k = 1}^n {{\sigma _k}}\text{，}$ (1)

 $S = \sqrt {\frac{1}{{n - 1}}\sum\limits_{k = 1}^n {{{\left( {{\sigma _k} - \bar \sigma } \right)}^2}} } \text{，}$ (2)

 ${\sigma _l} = {\sigma _{\max }} - {\sigma _{\min }}\text{，}$ (3)

1.2 分布特征参数（偏度系数、峰度系数）

 ${g_1} = \sqrt {\frac{1}{{6n}}} \sum\limits_{k = 1}^n {{{\left( {\frac{{{\sigma _k} - \bar \sigma }}{S}} \right)}^3}} \text{，}$ (4)

 ${g_2} = \sqrt {\frac{1}{{24n}}} \left[ {\sum\limits_{k = 1}^n {{{\left( {\frac{{{\sigma _k} - \bar \sigma }}{S}} \right)}^4}} - 3n} \right]\text{。}$ (5)
1.3 分布分析（概率密度函数、累计分布函数）

 $PDF\left( \sigma \right) = P\left( {{\sigma _0} \leqslant \sigma \leqslant {\sigma _0} + {\rm{d}}\sigma } \right) = \int^{{\sigma _0} + {\rm{d}}\sigma }_{{\sigma _0}} {p\left( \sigma \right)} {\rm{d}}\sigma \text{。}$ (6)

 $PDF\left( {n\Delta } \right) = p\left( {n\Delta } \right) = \frac{{{I_n}}}{{{J_N}}}\text{。}$ (7)

 $CDF\left( \sigma \right) = \int^\sigma_{ - \infty } {PDF\left( \sigma \right)} {\rm{d}}\sigma \text{。}$ (8)

 $CDF\left( {n\Delta } \right) = p\left( {n\Delta } \right) = \frac{{{J_n}}}{{{J_N}}}\text{。}$ (9)
1.4 百分概率值分布（10%概率值，50%概率值和90%概率值）

50%概率值对应的累计分布函数 $CDF\left( {\sigma {}_{50\% }} \right)$ 如式（10）所示。

 $CDF\left( {\sigma_{50\text{\%}}} \right) = \int_{ - \infty }^{\sigma_{50\text{\%}} }{PDF\left( \sigma \right)} d\sigma = 50\text{\%}\text{。}$ (10)

 $\sigma {}_{50\text{\%} } = \left[ {\sigma {}_{\min } + \left( {m - 1} \right)\Delta } \right] + \frac{\Delta }{{{I_m}}}\left( {0.5{J_M} - {J_{M - 1}}} \right)\text{。}$ (11)

 $CDF\left( {\sigma {}_{10\text{\%} }} \right) = \int_{ - \infty }^{\sigma {}_{10\text{\%} }} {PDF\left( \sigma \right)} {\rm{d}}\sigma = 10\text{\%} \text{，}$ (12)
 $CDF\left( {\sigma {}_{90\text{\%} }} \right) = \int_{ - \infty }^{\sigma {}_{90\text{\%} }} {PDF\left( \sigma \right)} {\rm{d}}\sigma = 90\text{\%} \text{。}$ (13)

2 基于BP神经网络的船舶目标识别方法 2.1 BP神经网络

BP网络是一种根据误差值反向传递，运用逆传播算法来对神经网络进行训练的多层前馈型网络。这种网络模型有一个最大的有点，就是它不需要事先知道相关的映射函数方程，就能学习和处理信息。这种网络模型主要是基于最速下降法的原理，利用反向信息传播来对网络中的突触权重和阈值进行自行调整，最终使所得误差的平方和最小。

BP网络可分为正向信息传递和误差值反向传递传递2部分。具体过程如图1所示。外界信息传递给输入层，而后输入层将外界信息传递给中间隐含层，这些信息在各隐含层经过处理后，最终传递给输出层，输出层则将这个结果输出。

 图 1 三层BP神经网络示意图 Fig. 1 Diagram of BP Neural Networks with three layers

2.2 基于BP神经网络的船舶目标识别

 图 2 基于RCS统计特征的船舶目标识别框图 Fig. 2 The diagram of ship target identification based on statistical characteristics from RCS
3 数据分析验证

 图 3 某船的RCS数据 Fig. 3 The RCS data of a ship

4 结　语

 [1] 王春花, 康猛, 郝明. 基于轮廓像的舰船目标识别方法研究[J].现代雷达,2012,34(4):45–47,52 http://www.cqvip.com/QK/93693X/201204/41695932.html [2] 赵安军,牛威,郭雷. 基于RCS的空间目标特征提取与识别研究[C]//Proceedings of the 27th Chinese Control Conference, 2008. [3] 隋栋训, 童创明, 王童, 等. 基于RCS序列的助推段弹道导弹识别[J].火力与指挥控制, 2016: 42(3): 15–18. [4] 张建强, 高世家, 赵霁红. 舰船RCS特征提取与GA-BP神经网络分类研究[J]. 船舶科学技术,2016,38(2):125–130. http://www.cqvip.com/QK/94984X/201602/668259882.html [5] 高隽. 人工神经网络原理及仿真实例[M]. 北京: 机械工业出版社, 2003.