Research On Intelligent Selection Of Road Network Automatic Generalization Based On Kernel-based Machine Learning

Road is an essential feature of the map,and the selection of road network is a fundamental operation in automated generalization.The problem has not been properly resolved as road network is very complicated and there are so many factors to consider in selection.Moreover,many factors are not explicit,and it is difficult to use the established model to express the importance of the road,which make it difficult to achieve good automatic selection in the conventional way.So in theory,intelligent approaches should be used in the problem of automatic selection of road network.Intelligent selection of road network is actually a simulation of expert selection process,in which establishing knowledge system,designing intelligent method for road selection and effective learning is the key to the research.So this paper focus on the following three aspects primarily:(1)Establish the knowledge system.Knowledge system is the basis of road selection.Setting up reasonable and effective selection parameter system is the primary task of road network selection.(2)Aautomatic acquisition of parameters.Based on the general road network data,this paper did not depend on the rigorous semantic information,and constructed a series of topological parameters according to the topology structure,and studied the automatic acquisition algorithm of the parameter value..(3)Design kernel-based machine model and do simulations.The design of kernel-based machine include selection of kernel function and optimization of its parameters.Using the designed kernel machine model,studying from training samples,and then experiment using varied road data is carried out.Optimizing and adjusting should be conducted continuously until the selection results meet the requirements.This paper tasks selection of road network as the research subject.According to the characteristics of the small scale road network that lacks basic attributes and semantic information,a series of topological parameters are built automatically as the basis of selection based on its existing property and topological structure,and a knowledge system which not rely on semantic information has been established.Combined with outstanding research achievements in the field of machine learning,using the powerful nonlinear mapping ability of the Kernel-based Machine,automatic selection is implemented through learning and training of samples.And experiments with various type of road network were conducted,which show that the method in this paper for selection of road network is feasible and effective,especially for radial and grid road network.