Automatic method to predict visual pleasantness and unpleasantness of streetscapes and identify key microscale components for improving pedestrian environments

Despite advances in computer vision-based streetscape evaluation, studies often overlook the influence of diverse microscale components and attributes like materials and combinations. This paper presents an automatic method to predict the visual quality of streetscape images from a pedestrian perspective, focusing on pleasantness and unpleasantness. Key components and combinations affecting this quality are identified. A dataset of 5000 streetscape images was developed, each labeled with 50 survey responses and component data. The image-based model outperformed previous approaches using both image and non-image inputs. The components contributing to pleasantness–unpleasantness were identified through Shapley-Additive-exPlanation analysis. Results showed that green space, traffic elements, pedestrian amenities, and street materials impact visual quality with varying combination effects. This study advances urban evaluation by developing an automatic method to predict streetscape quality and analyze microscale components. The findings contribute to practical urban improvements and facilitate more informed, effective decision-making in planning, design, and stakeholder engagement.