Measuring Temporal Evolution of Nationwide Urban Physical Disorder: An Approach Combining Time-Series Street View Imagery with Deep Learning

Urban physical disorder (UPD) is used to describe urban landscapes that are characterized by significant decay and deterioration. The phenomenon of UPD is also undergoing a process of evolution in conjunction with the developments that are taking place in urban areas. Measuring the evolution of UPD remains a challenge, however. This study innovatively employed time-series street view images and deep learning to analyze the temporal evolution of UPD on a nationwide scale, encompassing both overall levels and diverse manifestations. A total of 20 million street view images were collected in China from 2013 to 2022. The YOLOv8 object detection model was trained to accurately identify fourteen UPD elements. Subsequently, each element was assigned a weight to reflect the overall level of UPD. A clustering analysis based on graph neural networks identified four distinct manifestations, which were found to vary considerably across cities: good quality, vacancy and decay, under construction, and poor maintenance. The findings indicate a decrease in the overall level of UPD in China, with the primary issue shifting from poor maintenance to vacancy and decay. Economic growth correlates with overall improvements in UPD, whereas cities experiencing population decline tend to have more vacancies. The approach to measuring the evolution of UPD allows for a more nuanced understanding of the phenomenon and facilitates the prediction of urban space quality challenges, hence assisting urban planners in devising specific strategies for improvement.