Automatic Generation Of Urban Road Network Based On Deep Learning

For a long time,the road network was planned and designed under the participation of professional planners.It involves many processes including data preparation,field investigation,preliminary design and plan modification,which leads to the long design cycle and high design cost.Besides,the design quality is heavily dependent on the designer’s personal professional skills.With the development of artificial intelligence,especially in deep learning,the computer aided design of urban road network has become possible.The application of deep learning can not only reduce the workload of the planning staffm but also can reduce the design cost and improve the design efficiency,which are of great value in both academic research and applications.In recent years,researches have been carried out on the automatic modeling of road network,while these researches essentially transform the problem into other programmable matters.For example,L system uses the replacement rules of characters to simulate the generation of road network;in graph theory based methods,the planning of road network is transformed into the optimal path finding in graph theory.In that way,the problems in road network modeling are avoided.This paper creatively applies deep learning into the automatic modeling of urban road network and propose a deep neural network based model which is used to generate road network.The model,inspired by adversarial generative networks,takes road planning knowledge into account and be able to generate road network easily and rapidly.The model is trained on massive real road network data.By capturing the data distribution of training road samples,the model can ultimately understand the meaning of road network planning.The trained model can be used to generate real and reasonable road networks under specific constraints.In addition,considering that the training of the model requires a large amount of data,we also propose an algorithm used to automatically acquire road network samples.The algorithm uses the application programming interface provided by Baidu Map and have successfully collected the urban road network data of nationwide major cities.Meanwhile,the algorithm can also meet the requirement of automatic collection of massive data under other application scenarios.In this paper,two experiments are designed to evaluate the performance of the model in classification and generation of road network.From the results,we found that the accuracy of classification is high,the qualities of generated samples are fair and the road planning is reasonable,which achieved the intended training target.