Building function classification in Nanjing, China, using deep learning
Yongyang Xu, Zhanjun He, Xuejing Xie, Zhong Xie, Jing Luo, Hong Xie
Abstract
Abstract The functional classification of buildings is important for creating and managing urban zones and assisting government departments. Existing building function classification methods are mainly designed for remote sensing imagery or zones in vector maps. These methods cannot be used for the single buildings in large‐scale vector maps. In this study, a learning strategy for multiple features and context information is developed to detect a single building function in a vector map. First, multiple features are extracted for each building based on local and regional structures. Then, a graph convolutional network, GraphSAGE, is introduced to analyze the modeled graph and building footprint features through supervised learning. Experiments show that the framework can learn local and contextual building information with the ability to distinguish different building functions. When classifying the building function, the proposed method performed better than other machine learning methods, such as random forest and support vector machines.