Abstract: Livable streets play an important role in the construction of urban emergency facilities and disaster resilience, however, limited by the means of image analysis and the ability of data acquisition, existing research on street safety resilience is mainly based on large-scale qualitative, and it is difficult to carry out large-scale quantitative analysis based on the street's realistic visual environment. In this study, we utilize rich, easily accessible and ever-growing streetscape images as data sources, construct structural equation models and semantic segmentation methods for streetscape images, and utilize machine learning techniques for deep learning to pixelate the semantics of streetscape images of livable streets, and process selected segmented images with the help of the python code language, to identify and quantify the key environmental elements and indexes of livable streets, from pre-disaster The evaluation system is constructed in three dimensions of safety resilience planning and governance.

Living Street, Safety Toughness, Structural Equations, Street Images, Semantic Segmentation, Optimization Strategies