第 34卷 第 9期 控 制 与 决 策 Vol.34 No.9
2019年 9月 Control and Decision Sep. 2019
文章编号: 1001-0920(2019)09-1840-07 DOI: 10.13195/j.kzyjc.2018.1563
基于非规则标识点过程的遥感图像
零星地物目标几何特征提取
张洪云, 赵泉华
†
, 李 玉
(辽宁工程技术大学 测绘与地理科学学院,辽宁 阜新 123000)
摘 要: 针对遥感图像中零星目标几何特征的非规则性, 提出基于非规则标识点过程的遥感图像零星目标几何特
征提取方法. 首先, 将图像分为目标类和背景类, 利用非规则标识点过程建立目标的分布和几何特征, 其中非规则
标识采用一系列由节点连接起来的多边形定义而成,用以拟合目标几何形状,标识点用以确定目标位置; 然后分别
利用多值高斯分布和 KL(Kullback Leibler) 散度定义特征场能量函数和异质性能量函数, 形成全局目标提取能量
函数, 应用非约束吉布斯表达式将全局能量函数转换为概率分布函数, 并在最大化概率分布函数准则下设计合适
的 M-H(Metropolis-Hastings) 采样算法, 获得最优目标提取结果; 最后, 采用所提出方法对遥感图像零星目标进行
提取,由实验结果可以看出,所提出方法不仅能准确定位各零星目标,而且可以对其几何形状进行精确拟合.
关键词: KL散度;非规则标识点过程;目标提取;M-H算法
中图分类号: TP319 文献标志码: A
Geometric feature extraction of scattered targets from remote sensing
images based on irregular mark point process
ZHANG Hong-yun, ZHAO Quan-hua
†
, LI Yu
(School of Geomatics,Liaoning Technical University,Fuxin 123000,China)
Abstract: In view of the irregularity of the geometric features of the scattered targets in remote sensing images, a
geometric feature extraction method for remote sensing images based on irregular mark point process is proposed. Firstly,
the image is divided into target and background, and the distribution and geometric features of the target are established
by using the irregular marking point process. The random point are used to determine the target position, and the marks
associating individual points are defined with a set of nodes to depict the geometric shape of the target. Then, on this
basis, the multi-Gaussian distribution and Kullback-Leibler divergence are used to define the energy function and the
heterogeneous energy function, and form the global energy function. The global energy function is converted to the
probability distribution function by using the unconstrained Gibbs function, and the appropriate Metropolis-Hastings
sampling algorithm is designed under the maximization probability distribution function criterion. The optimal target
extraction result is obtained. Finally, the proposed method is used to extract the scattered targets from remote sensing
images, it can be seen from the experimental results that the proposed method achieves better experimental results. The
traditional mark point process method is extended to the irregular mark point process. According to the experimental
results of remote sensing image extraction, it can be seen that the method can be more accurate to fit the scattered targets.
Keywords: KL divergence;irregular mark point process;target extraction;Metropolis-Hastings algor ithm
0 引
图像的主要信息一般集中在几个少数的关键
区域 (如道路、河流、水库、小岛等) 内, 当这些区域分
布不均匀且较为零散时,称之为遥感图像中零星地物
目标
[1]
. 这些零星目标的几何特征表现为条形或面域
多边形等, 可以准确地反映目标地物的形状轮廓、长
宽尺寸及目标像素的分布等特点
[2]
,在目标检测和识
别过程中具有重要应用, 因此, 如何准确提取遥感图
像中零星目标的几何特征成为学者们重点研究的内
容之一
[3-4]
. 常用的遥感图像地物目标几何特征提取
方法有基于特征判别的方法
[5]
、基于分类学习的方
法
[6]
、基于 Hough 投票的方法
[7]
以及基于目标模型的
收稿日期: 2018-11-13;修回日期: 2018-03-04.
基金项目: 国家自然科学基金项目 (41271435, 41301479);辽宁省自然科学基金项目 (2015020090).
责任编委: 张国山.
†
通讯作者. E-mail: zqhlby@163.com.