矢量道路数据的自动匹配与变化检测研究

发布时间：2018-05-01 20:28

本文选题：道路网 + 同名匹配　；参考：《南京师范大学》2017年硕士论文

【摘要】：道路网空间数据是基础地理数据库中的重要组成部分,也是导航应用、灾害救援、物流交通等专题数据的重要内容。因此道路数据的现势性直接决定这些应用能否准确有效。为促进经济快速发展,国家对基础建设,特别是各类道路网络的建设,投入大量资金,我国道路网建设如火如茶,可谓日新月异。道路实体的变化也促使各类空间数据库中的道路数据也必须及时更新,只有这样才能保证空间数据库的现势性。目前,增量更新是数据库更新的重要方式之一,是保持数据库现势性的重要手段。而道路要素的同名匹配和变化检测是道路网增量更新过程中的两个关键流程,其中同名要素匹配是实现增量更新的基础,只有先完成要素匹配才能在此基础之上检测是否发生变化;而变化检测是增量更新的前提,因为,只有实现对变化区域变化要素的检测和提取,才能进行增量更新。针对以上需求,本文引入径向基函数网络理论和决策树理论对道路要素的同名匹配和道路网变化检测与分类进行研究,主要成果如下:(1)对道路网要素的自动匹配和变化检测两个方面的国内外研究现状进行了总结和分析,在归纳当前的研究方法的基础之上探讨了其中存在的一些问题。依据道路网的变化规律,归纳了道路变化特征因子和道路变化类型,为后面的道路网自动匹配和变化检测打下了理论基础。(2)提出了基于径向基函数网络的多特征因子路网匹配方法。本文综合利用道路网中路段的长度、距离、形状、方向等几何特征的相似度和结点的拓扑特征的相似度等5个空间特征相似度指标对多源道路网进行相似度判断。为解决各个相似度指标在匹配中的权重分配问题,引入径向基函数网络理论,并对经典径向基函数进行改进,改进后的径向基函数网络顾及了不同路网相似度指标在路网匹配中所起作用不同这一特点,使径向基函数具有各向异性特征。在神经网络输出层引入sigmoid函数,对匹配结果值作归一化处理,从而实现道路网的可靠匹配。通过与常用的BP神经网络在道路网匹配中的效果进行比较,实验证明径向基函数网络在样本训练和路网匹配时效率更高,匹配准确率也更高。(3)提出了基于决策树的道路网变化检测与分类方法。设定路长、路型、方向、结点度以及属性等5个道路变化特征因子作为决策树的特征,改进传统决策树生成过程中基于信息增益选择特征的方法,利用道路变化特征影响力算法快速计算样本数据的道路变化特征的影响力值,基于特征影响力值排序来选择决策树特征,生成道路变化分类的决策树,进而完成道路网变化的检测和变化类型分类。(4)在本文理论研究成果的基础之上,设计并开发了道路网变化检测与分类原型系统。该系统可以实现道路网数据的管理、简单的数据预处理、道路网同名匹配、道路网变化检测以及道路数据的变化信息查询、属性查询和空间查询等功能。
[Abstract]:The road network spatial data is an important part of the basic geographic database, and it is also an important part of the special data of navigation applications, disaster relief, logistics and transportation. Therefore, the potential of road data directly determines whether these applications can be accurate and effective. In order to promote the rapid economic development, the national infrastructure, especially the various road networks, can be used to promote the rapid economic development. Building and investing a lot of money, the construction of road network in China is like tea, which is changing rapidly. The change of road entity also prompt the road data in all kinds of spatial databases to be updated in time. Only in this way can we guarantee the potential of spatial database. At present, incremental updating is one of the important ways to update the database, and it is to maintain the database. The same name matching and change detection of road elements are the two key processes in the process of incremental updating of the road network, in which the matching of the same name elements is the basis for incremental updating. Only by completing the matching of elements before the detection is based on this, the change detection is the prerequisite for incremental updating. In order to carry out incremental updating only by detecting and extracting the factors of changing regional changes, this paper introduces radial basis function network theory and decision tree theory to study the homonym matching of road elements and the detection and classification of road network changes. The main achievements are as follows: (1) automatic matching of road network elements and the main results are as follows. The current research situation at home and abroad in two aspects of change detection is summarized and analyzed. On the basis of summarizing the current research methods, some problems are discussed. According to the change law of road network, the characteristic factors of road change and the type of road change are summed up, and the automatic matching and change detection of the road network are laid down. The theoretical basis. (2) a multi characteristic factor road network matching method based on radial basis function network is proposed. In this paper, the similarity degree between the length, distance, shape, direction and other geometric features such as the length, distance, shape, direction and other geometric features of the road network is used to judge the similarity degree of the multi source road network. In order to solve the weight allocation problem in the matching, the radial basis function network theory is introduced, and the classical radial basis function is improved. The improved radial basis function network takes into account the different functions of the similarity index of different road networks in the road network matching, and makes the radial basis function have the anisotropic characteristics. The sigmoid function is introduced through the network output layer, and the matching results are normalized to achieve a reliable match of the road network. By comparing the effects of the common BP neural network in the road network matching, the experimental results show that the radial basis function network is more efficient and more accurate in the sample training and road network matching. (3) The method of road network change detection and classification based on decision tree is proposed. 5 road change features are set as the characteristics of the decision tree, which are the path length, the road pattern, the direction, the node degree and the attribute, and the method based on the information gain selection feature is improved in the traditional decision tree generation process. The method is used to quickly calculate the sample with the road change feature influence algorithm. The influence value of the road change characteristics of the data, based on the characteristic influence value sorting to select the decision tree characteristics, generate the decision tree of the road change classification, and then complete the road network change detection and the change type classification. (4) on the basis of the theoretical research results of this paper, the road network change detection and classification prototype system is set up and developed. The system can realize the management of road network data, simple data preprocessing, road network homonym matching, road network change detection, road data change information query, attribute query and spatial query.

【学位授予单位】：南京师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P208

【参考文献】