测绘学报

• 学术论文 • 上一篇    下一篇

基于C4.5算法的道路网网格模式识别

田晶1,艾廷华2,丁绍军3   

  1. 1. 武汉大学资源与环境科学学院
    2. 武汉大学资环学院
    3. 武汉大学 资源与环境科学学院
  • 收稿日期:2010-12-13 修回日期:2011-01-30 出版日期:2012-02-25 发布日期:2012-02-25
  • 通讯作者: 田晶

Grid Pattern Recognition in Road Networks Based on C4.5 Algorithm

  • Received:2010-12-13 Revised:2011-01-30 Online:2012-02-25 Published:2012-02-25

摘要: 道路网模式的识别对于地图综合、数据匹配和空间分析具有重要意义。网格模式是道路网中的典型模式之一。本文提出一种基于C4.5算法的网格模式识别方法。该方法以道路网中的网眼多边形为基本单元,根据上下文关系将其标识为属于网格模式和不属于网格模式两类。首先采用形状参量和关系参量描述网眼多边形,然后,基于决策树C4.5算法分别对5维参量和3维参量构造分类器,运用10折交叉验证获得具有说服力的结果,其Kappa值分别为0.63和0.66,正确率分别为81.7%和82.9%,置信度90%的置信区间分别为[0.785, 0.846]和[0.797, 0.857]。在新数据上进行了识别效果的验证,结果表明该分类器可用于网格模式的识别。研究试图将传统模式识别和数据挖掘的理论方法应用于空间问题的解答中。

Abstract: The pattern recognition of spatial cluster has become a hot issue in the areas of geographical information sciences. Pattern recognition of road networks plays an important role in map generalization, data matching and spatial analysis. Grid pattern is one of the most typical patterns in road networks. A grid is characterized by a set of mostly parallel lines, which are crossed by a second set of parallel lines with roughly right angle. This paper proposes a method for grid pattern recognition based on C4.5 algorithm. Meshes in road networks can be classified as belonging to grid and not belonging to grid according to their context. Firstly, shape measures and relation measures are defined to characterize meshes in road networks. Secondly, two classifiers are trained using C4.5 algorithm based on five measures data and three measures data. A 10-fold cross validation process is applied in order to obtain a sounder result. Finally, the performance of the classifiers is evaluated by means of the Kappa index and the overall correct rate. The Kappa classification accuracy for five dimensions data and three dimensions data is 0.63 and 0.66. The overall correct rate is 81.7% and 82.9% for each. The confidence interval of 90% confidence is [0.785, 0.846] and [0.797, 0.857] respectively. The classifiers are tested by a new data set and the results show that the classifiers are valid in grid pattern recognition. This study tries to use theories and methods of traditional pattern recognition and data mining to solve the spatial issues.