Objective As global urban renewal advances toward high-quality development, human-scale perceptual landscapes have become central to urban regeneration. However, existing research inadequately addresses how different renewal pathways influence citizens’ perceptual experiences of urban environments. Meanwhile, accompanying the paradigm shift from “demolish-construct-preserve” to a balanced “preserve-renovate-demolish” approach, two typical renewal modes have gradually emerged: First, “demolition-reconstruction renewal,” which achieves spatial transformation through functional replacement and element restructuring, often resulting in structural changes to urban physical environments. Second, “preservation-renovation renewal,” which advances transformation through micro-updates such as building facade improvements and pedestrian environment optimization without altering existing landscape structures. Based on this context, our research objectives include: 1) Mapping the spatial distribution and temporal evolution of multi-dimensional visual perception in urban core functional areas, identifying which urban landscape types are consistently associated with higher or lower perceptual quality. 2) Examining whether differential associations exist between different renewal pathways and perceptual changes.

Methods This study focuses on Shanghai’s central activity zone. As one of the most structurally complex and rapidly evolving urban renewal areas in China, this zone provides an ideal context for examining diverse renewal pathways. The research constructs a comprehensive analytical framework combining temporal street view imagery analysis with machine learning techniques. Street view data were collected using systematic sampling at 50-meter intervals, capturing images from four cardinal directions at each observation point, with a temporal span from 2017 to 2022. The methodology integrates three core analytical components: First, urban landscape perception assessment utilizes the internationally recognized Place Pulse 2.0 dataset and ResNet50 network architecture to evaluate six perceptual dimensions: safety, vitality, affluence, beauty, interest, and pleasantness. Python-based ZENSVI packages enable automated perception prediction through pre-trained deep learning models. Second, urban physical landscape clustering employs semantic segmentation using GluonCV with DeepLab v3+ models trained on the ADE20K dataset. Based on semantic segmentation results, we constructed 11 core landscape characteristic indicators, including sky openness, green view ratio, traffic orientation index, building aspect ratio, and others. A K-means clustering approach based on the GraphSAGE (Graph Sample and Aggregate) graph neural network identifies five typical urban landscape types, incorporating spatial constraints through multi-scale distance thresholds and exponential distance decay weights. The GraphSAGE generates spatially-aware feature vectors by sampling and aggregating features from node neighborhoods, combining node attributes with neighborhood information. Third, the renewal mode impact analysis employs a two-step analytical framework: ANOVA tests examine perceptual score differences across different landscape clusters, establishing baseline relationships; chi-square tests examine the associations between renewal modes and perceptual change directions, while the net perceptual change index (NPCI) quantifies the net perceptual gain or loss of each renewal pathway.

Results The spatial distribution analysis reveals that visual perception quality in Shanghai’s central activity zone exhibits significant spatial clustering and differentiation characteristics. High-quality perception areas concentrate in urban core zones including the Hengshan-Fuxing Historic and Cultural Area, Nanjing West Road, People’s Square, and Xintiandi, which serve as mature commercial, cultural, and social centers with refined built environments, rich commercial vitality, and profound historical heritage. Conversely, low-quality perception areas vary by dimension, with “beauty” and “pleasantness” cold spots mainly concentrated in the Old City, North Sichuan Road, and Shanghai Railway Station vicinity, while “interest,” “vitality,” “safety,” and “affluence” cold spots appear significantly in Xuhui Riverside, Qiantan, and Houtan areas. The 2017–2022 perception change analysis demonstrates that urban renewal processes generate complex spatial characteristics, with perception improvement and deterioration areas displaying relatively discrete rather than concentrated distribution patterns, reflecting the point-based, micro-renewal dominant renewal mode. Overall, areas with perception improvements outnumber deteriorated areas, with pleasantness and interest showing particularly notable enhancement. ANOVA and Tukey’s HSD tests reveal statistically significant differences across all six perceptual dimensions among different landscape cluster types. Cluster 2 (green landscaped roads) and cluster 3 (mixed-function community streets) achieve the highest multi-dimensional perception evaluations, while cluster 1 (traffic-oriented arterials) and cluster 4 (open undeveloped spaces) receive the lowest perception scores across all dimensions. Effect sizes indicate that landscape types most significantly influence vitality and beauty perceptions, while their impact on affluence perception remains limited. Dynamic analysis through chi-square testing reveals that among six perceptual dimensions, only beauty and pleasantness changes show statistically significant associations with urban landscape renewal pathways (p<0.001). Net perceptual change index analysis demonstrates that preservation-renovation renewal in high-quality landscape features can achieve perception improvements potentially exceeding demolition-reconstruction renewal benefits. Notably, when other landscape types are updated to “green landscaped roads” and “mixed-function community streets,” they generally generate more positive perceptual changes.

Conclusion Based on temporal street view analysis from 2017 to 2022, this study demonstrates that the perceptual effects of urban renewal are not solely determined by the intensity of physical intervention. Static perceptual advantages of certain landscape types do not necessarily translate into proportional perceptual gains during renewal processes, revealing clear asymmetries across perceptual dimensions. In dense urban core contexts dominated by incremental renewal, preservation-renovation can generate perceptual benefits comparable to or greater than demolition-reconstruction in high-quality existing environments, while the latter remains more effective for improving low-quality spaces. By linking renewal pathways to dynamic perceptual change, this study provides an evidence-based framework to support differentiated, human-centered urban renewal strategies.