道路提取作为利用遥感影像海量信息的一个重要组成部分，具有广阔的应用
前景。随着遥感技术的发展，针对高分辨率遥感影像的道路提取是当前的研究热
点，具有重要的科学意义和实用价值。

本文给出了一种基于线特征的高分辨率遥感影像道路提取方法，在分析了高
分辨率遥感影像中道路上线特征的特点之后，提出了条带状直线段集合形式的道
路模型，并参照马尔视觉理论采用了先提取低层次的线特征，再识别出中层次表
示道路的线特征集合，最后合并简化概括为高层次道路的自底向上的基本思路。

本文首先讨论了图像预处理过程中的滤波重采样及图像分块处理操作，以应
对影像中的噪声、阶梯效应及遥感影像数据量大的问题；接着分析了线特征的实
质，在评价了已有的线特征提取方法后，以目前较先进方法为基础进行改进，给
出了适合本文的能够较快速提取较完备直线段特征的提取方法；然后对同属于道
路的直线段之间的拓扑关系进行分类，说明了通过计算对不同拓扑关系进行判别
的方法以及通过R 树实现的直线段最近邻查询的方法，给出了相邻直线段同属于
同一道路的可能性度量，并详细阐述了基于拓扑关系的道路线特征集合的识别过
程；接着对直线段集合分段，将各段简化为一条直线段，并用平滑折线拟合连接
得出道路中心线；最后通过实验和统计数据对方法进行了评价。

另外，在本课题的研究过程中，为查看和比较不同实验的结果需要进行大量
的重复性实验工作，针对此图像处理与模式识别中的常见问题开发了数据处理可
视化实验平台。本文附录介绍了该平台的系统定义、技术支持与操作流程，通过
类图详细说明了此平台的框架结构设计，对于图形拖拽实现修改、系统平台文件
的存储与读取等关键功能也给出了具体实现方法。

As an important component of using mass data of remote sensing images, road
extraction has a wide applicaton prospect. With the development of remote sensing
technology, road extraction from high resolution remote sensing images, which
relating to important scientific Significance and practical use, has become a hot topic.

A road extraction method from high resolution remote sensing images is
proposed in this paper. Banded sets filled with line segments are used as the road
model, after analysising the characteristic of line segments acquired from roads.
Referring to Marr's vision theory, an research thread that get line segments at low
level, recognize banded sets standing for road at middle level, and finally simplify
banded sets to road at high level, is used.

To treat noises, staircase effects and massive data, filtering, sub-sampling and
processing in blocks during the preprocessing stage are firstly discussed. The
substance is analysised then. After evaluating existing line segments extracting
methods, a line segments extracting method fitting this paper and based on the better
method at present is proposed. Next, topology between line segments on road is
classified and judged by computing, nearest neighbors query is realized by R-tree,
measurement method of probability that adjacent line segments belonging to one road
is presented, and recognation method for road banded sets based on topology is
elaborated. Then, the road banded sets is devided into small parts, and each part is
simplify by a line segment, which is used to connected by a polyline standing for the
road. At last, the whole method is evaluated through experiment and statistics.

Besides, In order to see and to compare results of different experiments large
amounts of repeating experimentation is needed during the research of this topic. To
treat problems of this kind in the research work of image processing and pattern
recognition, A visual experiment platform system for processing data is developed.
The system definition, technical support, as well as operating procedure of this
platform are explained. Framework design is stated using the class diagram, key
functions such as modifying graphics by draging and how to read and write files for
the platform are also illustrated in the appendix.