As one of explicit patterns widely existing in urban road network,grid patterns contain a large amount of abundant information about urban spatial pattern.Recognizing road grid patterns is also the key prerequisite for realizing automatic and intelligent map generalization.As existing methods of grid pattern recognition seldom consider multi-level mesh features and may be sensitive to the diversity of training samples,this paper proposes a novel approach for urban road grid pattern recogni-tion based on multi-level mesh features and VAE-PNN model.Firstly,the original road network data are simplified by sophis-ticated DP algorithm.Then,we design a set of multi-level mesh features to measure grid patterns,including internal orthogo-nal function,grid shape descriptions and neighborhood correlation indicators.After that,variational auto-encoder(VAE)is used to enhance the diversity and size of training samples.Finally,probabilistic neural network(PNN)model is adopted to identify road grid patterns.The experimental results show that considering multi-level mesh features can help to identify differ-ent grid types and various grid patterns and demonstrate that introducing VAE model achieves better performance on grid pat-tern recognition than not.
关键词
格网模式识别/多层次网眼特征/变分自编码器/概率神经网络
Key words
grid patterns recognition/multi-level mesh feature/variational auto encoder/probabilistic neural network
维普
万方数据