딥러닝 모델을 활용한 신도시와 구도시 간 도시 내 이미지 상대적 특성 분석 - 성남시를 대상으로

This study aims to quantitatively evaluate urban image scores using street view images and deep learning technology, focusing on analyzing the relative image characteristics of new and old districts in Seongnam City. The ResNet-152 model was trained using the Place Pulse 2.0 dataset, and the urban image scores of Seongnam City were measured across six indicators: Safety, Lively, Wealthy, Beautiful, Boring, and Depressing. The analysis revealed that Safety and Beautiful scores were higher in the old city, while Boring scores were higher in Bundang New Town. Additionally, Depressing scores were elevated in both the old city center and Bundang New Town, whereas Wealthy and Lively scores exhibited a more uniform distribution across the regions. Correlation analysis identified a negative relationship between Safety and Boring indicators. These findings clearly demonstrate the relative urban image characteristics between new and old districts. By proposing a rapid and standardized urban image evaluation methodology, this study contributes empirical data essential for urban policy formulation and the development of strategies for both new and old districts.