Evaluation of architectural colour comfort and harmony based on street view images: a case study of Caidian

Architectural colour significantly shapes urban visual identity and influences spatial perception and emotional experience. However, rapid urbanization and poor planning have led to chaotic, culturally disconnected colour schemes, weakening urban character and degrading residents’ environmental perception. Traditional field-based studies are limited in scale and objectivity, while advances in street view imagery and machine learning enable large-scale, human-centred, and quantitative analysis of urban colour. Motivated by these considerations, this study initially collects Street View Images (SVI) using Python-based deep learning, with architectural colours extracted through semantic segmentation and k-means clustering. Subsequently, an evaluation framework integrating colour perception analysis and DBSCAN clustering – a machine learning – is established to assess streetscape visual comfort and the colour harmony of building clusters. Based on the findings, a multi-scale strategy for architectural colour management is proposed. This study supports the refinement of urban design strategies, contributing to the continuous improvement of environmental quality.