Recognition of Grid Pattern in Street Network Using Self-organizing Maps

Pattern recognition has been a hot issue in the field of map generalization.Grid pattern can be regarded as consisting of cluster of meshes,which have similar properties in location,size,shape and relationship.We present an approach to the recognition of grid pattern in street network using self-organizing maps.Firstly,eight parameters,i.e.centroid,area,rectangularity,elongation,having parallel side,side number,number of 1st order neighbor,mean rectangularity are selected to describe meshes.Each mesh constitutes a vector in attribute space.These vectors then are used to train a SOM.The neurons of a SOM correspond to a set of meshes with similar properties.U-matrix is used to visualizing the SOM.Grid pattern recognition is based on the clusters identified by the SOM.Two experiments are conducted.The results show the proposed approach is valid in mining the grid pattern in irregular street networks.The limitation and future work are discussed.