Evaluation of spatial visual perception of streets based on deep learning and spatial syntax

Street visual quality improvement plays an important role in urban development. An important direction for street quality research lies in accurately perceiving the spatial quality of urban streets and exploring the connection with street constituents. This study applies deep learning to extract visual elements from street view images and uses a human-machine adversarial model to rate them across six dimensions (beautiful, wealthy, safety, lively, depressing, and boring). Through spatial visualization of street quality, overlay analysis with network accessibility, and multiple linear regression, it examines the correlations between street space quality and its constituent elements. The results indicate that the streets within the study area scored highly on the dimensions of beautiful and lively, this is attributed to the reasonable greening construction which characterized by the good layout and density of greenery. Such greening not only enhances aesthetics but also provides environmental benefits. Additionally, the orderly street layout reflects well-organized spatial arrangements of street elements, such as pathways and building facades. Positive visual perception such as beautiful, wealthy, safety, lively is positively correlated with plants and pedestrians, and negatively correlated with walls. It is important to address the distinct types of streets in urban planning, including high-quality/accessibility streets found in urban centers, high-quality/low-accessibility streets at district junctions with sparse networks, low-quality/high-accessibility streets in the southwestern center and Low-quality/accessibility peripheral areas characterized by outdated buildings. Strategies should prioritize improvements in street function, greening, building interfaces, and pedestrian connectivity. These measures will help enhance the overall spatial quality and vitality of the area. In summary, the findings provide data support for more precise urban street improvements and offer a reference for human-centered urban planning research.