Quantifying perceived decline through deep learning predictions and analyzing association with objective measures: A case study of Seoul, S. Korea

Abstract

The deterioration of commercial districts in Seoul, South Korea, is intensifying due to competition among the self-employed, COVID-19 impact, and declining offline consumption. Recognizing the importance of commercial districts in urban vitality, reinvigorating these areas is crucial. The initial step in such efforts would be assessing the extent of the decline. This study investigates the possibility of identifying perceived decline using computer vision and examines its relationship with objective decline measures. First, a deep learning model was developed to predict perceived decline scores (PDS) using a self-developed training dataset generated from street view imagery-based website surveys (n = 3393). Then, spatial regression analyses examined correlations between objective measures and predicted PDS. The results showed the following: (1) uneven PDS distribution across regions; (2) PDS increased significantly with decreased commercial building ratio, increased old building ratio, classification as a major commercial area (main street), classification as a traditional market, decreased sales per store, decreased greenery, and decreased new permits per area. This study contributes to existing literature by empirically demonstrating the association between subjective perception—mostly visual attributes—and objective measures on urban decline, enabling local governments to conduct preliminary investigations into several aspects of area decline cost-effectively.