Urban Street-Scene Perception and Renewal Strategies Powered by Vision–Language Models

Urban renewal research has long relied on expert-led assessments and fragmented indicators, yet lacks scalable, perception-aware frameworks that can translate street-level conditions into interpretable renewal strategies. To bridge these gaps, this study proposes a vision–language model (VLM) based method to identify the potentially renewable areas across the Hongshan Central District of Urumqi, China. Specifically, we collected 4,215 panoramas and used multiple VLMs to measure six perceptual scores (i.e., safety, liveliness, beauty, wealthiness, depressiveness, and boringness) together with textual descriptions. The best-performing model, selected by correlation with a 500-respondent perception survey, was used as the final analysis to identify the renewal area. Then, we conducted spatial statistics and text mining (eight semantic themes) to reveal the spatial patterns and semantic topics for proposing renewal strategies. The results show that: 1) VLMs have a high consistency with humans in evaluating the spatial perception of six dimensions; 2) four renewal priority tiers were identified, with high-score areas concentrated on Tianshan District Government Residential Quarter, Mashi Community, Heping South Road, etc.; and 3) Semantically, low-score areas such as Hongshan Road, Binhe Middle Road, Wuxing South Road, Huhuo Line, etc. emphasize infrastructure, safety, street level and order. We conclude that VLMs add value not only via scalable assessment but also through explanatory language evidence that directly supports tiered renewal and public communication. This work provides a data-driven and interpretable evaluation framework for urban renewal decision-making, facilitating precision-oriented and intelligent regional urban regeneration.