Study Of Complex Spatial Relation Model Based On Decision Trees

The reasearch of the Spatial-Temporal reasoning originated from the early 1970s.It use techniques and methods of artificial intelligence in modeling,description and express spatial objects.at the same time deal with and analysis relationships of spatial objects.The reasearch of spatial-temporal has made great progress,it has an extremely important theoretical and pratical significance in atrificial intelligence.But there exists some questions also.for example,the research is not enough in Complex spatial relationships of objects and integration of various of spatial reasoning.The research of spatial-temporal DATA Mining based on spatial-temporal and multi-relational data mining is not enough.Most of existing spatial-temporal and spatial reasoning methods difficult to solve practical problems.As the increasing of data between objects with complexity spatial relationships. It has become more and more important to search for specific and useful information.in particular difficult to formalize modeling for complexity of spatial relationships exist in data.Topology relationships between spatial objects are the most basic relationships between objects.the relationships of topology constitute the basic method of spatial-temporal reasoning.but the traditional model of tipological relationships between spatial objects is difficult to description all the relationships between spatial objects.However,in pratical applications.always need to give specific representation an dreasoning for spatial relationships of objects.although there exists many metric attributes of objects,the extent of impact of metric attributes are different.One or some kinds of metric attributes may directly deterimined the relationships of objects,Some of metric attribtes may didn't have affect on relationships between objects.this lead to difficult to formalize modeling between spatial relationships exist in data set.In response to these questions.In this paper based on the method of relationship between spatial-temporal and multi-realtion data mining,study in-depth of metric attributes between objects,give the metric attribute model of directed line,at the same time give the corresponding model calculation of the index value.In this paper,based on metric of simple spatial relationships,analysis the spatial relationships of directed lines,proposed the complex spatial relationships model to deal with object relationships.in the aspect of complex spatial relationships have gain beneficial explore.in this paper build the classifier system based on decision trees in data mining,this system applied in the formal model spatial relationships,then finished the procedure of formal model,give the formal express of spatial relationships.In this paper,classifier system build based on topology relationships,metric attributes and description of metric attributes.the important parts of this paper is spatial analysis of the object ,in particular spatial relationships of directed line.the specific work of this paper as follows:1.is paper briefly introduce the knowledge of related research and theory.Including topological relations model of the n-intersection model and 9-intersection model.detailed description relationships between line-line topology relationships.Line-Line topological relationships build based on point and point in the space,space points and line segments.this as the basic build a new method of topology relationships.This method include 56 kinds of line-line topology relationships.give the principal of decision trees,include the purpose of attribute analysis,procedure of attributes and method of attribute analysis. Give the information theory and entropy,informatin gain and gain function in detail.2.this paper,study topology relationships of sptial-temporal in-depth.analysis 9-intersection model in detail,analysis topology relationships of line-line in summary.Because can not represent complex spatial relationships of directed line objects only use topology model,so this paper propose the metric attributes between directed lines.Metric model include 4 kinds.1.index model of direction for directed line objects.direction index include local angle and global angle.2.index model of split for directed line.it includes intersect split,alongness split and interior traversal split.3.index model of distance for directed line.it includes shortest distance ratio index,longest distance ratio index and shortest-longest distance ratio index.4.index model of overlay ratio for directed lines.Use these model as fundation analysis complex spatial relationships of directed line objects,give the method to calculus index value,formalize the data-set combined by metric attributes.use these as foundation formal modeling complex spatial relations.but only have the metric attributes can not formal model which hide in the data-set.so in this paper use the method of decision trees as the foundation,build the classifier system,use this system mining the data-set,then give the formalize represent of complex spatial relationships.3.e main modules of classifier system is mining spatial data-set.use the result of mining formalize model of spatial relationship.This system read the data-set from file,use the method of decision trees analysis spatial relationships,draw the result of classification for spatial relationships.from the result summarize the tipology and metric attributes which affect spatial relationships.Alough ID3 algorithm of decision trees is the most basic and influential,it can not deal with numeric attributes.howerve most of attributes of spatial relationships are numeric,so this paper use the algorithm of C4.5 mining data-set.The main idea of this algoritm is compute the information gain for every node.select the node which has the most value of information gain as the split node.Because of there may be exist the impurity data in data-set,so we need to prune the decision tree,the method of prune include pre-prune and post-prune,this system use the method of post-prune.from the decision trees obtained from the result of classification,we obtain the graphic represent of decision trees.at the end use IF…Then…principal represent the main topology and metric attribute which affect the complex spatial relationships,implementation the formal model of spatial objects.The index model of metric attribute propose for directed line objects,the classifier system build based on decision trees have important significance on formalize modeling for complex spatial relationships hide in data-set.but this method can not memory the sequence of spatial relation.so the next step is find the method which can memory the sequence of spatial relationship.formal modeling the complex spatial relationships of objects by happen sequence of spatial relationships.