Building Type Classification from Street-view Imagery using Convolutional Neural Networks

Articles and reports: 18-001-X2021003


Micro-level information on buildings and physical infrastructure is increasing in relevance to social, economic and environmental statistical programs. Alternative data sources and advanced analytical methods can be used to generate some of this information. This paper presents how multiple convolutional neural networks (CNNs) are finetuned to classify buildings into different types (e.g., house, apartment, industrial) using their street-view images. The CNNs use the structure of the façade in the building’s image for classification. Multiple state-of-the-art CNNs are finetuned to accomplish the classification task. The trained models provide a proof of concept and show that CNNs can be used to classify buildings using their street-view imagery. The training and validation performance of the trained CNNs are measured. Furthermore, the trained CNNs are evaluated on a separate test set of street-view imagery. This approach can be used to augment the information available on openly accessible databases, such as the Open Database of Buildings.

Issue Number: 2021003
Author(s): Al-Habashna, Ala'a
FormatRelease dateMore information
HTMLJanuary 21, 2022
PDFJanuary 21, 2022