Deep Convolutional Autoencoder Model for Urban Land Use Classification using Mobile Device Data

Location-based information acquired from mobile phone data provides insights into the interactions between individuals and their urban environment such as urban land uses. Using mobile phone data and the deep convolutional autoencoder (DCAE) model, the study aims to recognize citywide land uses in a more accurate and timely manner, which can serve as important reference for urban planning practices. The proposed technique uses unsupervised learning based on DCAE model and automatically recognizes land uses in urban areas by clustering geographical regions with similar features in curve patterns and volumes of mobile phone calls in a time series. A case study of Wuhan city is presented, and experimental results are validated using land use information provided by the city planning departments. The study concludes that our method can more effectively recognize and classify land use types in the case city compared with previous methods.