CLASSIFICATION OF URBAN MORPHOLOGY WITH DEEP LEARNING: A CASE STUDY OF ITS APPLICATION ON URBAN VITALITY PREDICTION

Abstract

Urban morphology studies the physical form of cities. There is a prevailing trend to study urban morphology quantitatively thanks to the growing accessibility to all kinds of spatial big data. The quantitative methods up to now managed to characterize urban morphology with morphological indices describing density, proportion, and mixture but never attempted to measure the feature from human’s visual and intuitive perception. This study took the first step to alleviate the gap by proposing a deep learning-based technique to classify road networks into gridiron, organic, radial, and no pattern, and embed this feature into probabilities. It was achieved by the classification of Colored Road Hierarchy Diagram (CRHD), which was introduced and generated by the author based on Python and OpenStreetMap (OSM) datasets. The probabilities could be appended to traditional morphological index collection as an augmentation of human perception. The technique was realized by ResNet-34 and was able to reach an overall classification accuracy of 0.86. Nine representative cities were selected as the study areas. Latent subgroups among the cities were uncovered through a simple clustering on the percentage of each road network category. The aforementioned method is independent and could be applied in many morphology-related studies. In this paper, the effectiveness of the human perception augmentation was examined by a case study of urban vitality prediction. The measurement of urban vitality incorporated multiple facets including built environment vibrancy, human activity density, nighttime light brightness, population density, and tourism vibrancy to mitigate the biases caused by every single indicator. An advanced machine learning model, LightGBM, was for the first time designated to establish the relationship between morphological indices and vitality indicators. A Positive effect of human perception augmentation was detected in the comparative experiment of baseline model and augmented model. The study also revealed the differentiation among road network categories on their upper limits of vitality incubation. Its causation was analyzed with a theory of interactive facade. Suggestion on boosting urban vitality was offered from the angle of the selection of road network category. To conclude, this study expanded the toolkit of quantitative urban morphology study with new techniques. More urban morphology studies associated with these techniques could be conducted in the future.