Urban vitality provides a human-activity-based lens for evaluating how existing urban land supports everyday use and is therefore relevant to sustainable urban regeneration in compact cities. This study examines block-level activity patterns and their spatial correlates in the Macau Peninsula using Baidu Heat Map raster imagery from two sampled Mondays and two sampled Saturdays in summer 2020, together with POIs, road networks, building data, vegetation, waterfront conditions, public transport, and socioeconomic indicators for 159 blocks. We treat the Baidu-derived measure as relative activity intensity rather than an absolute population count or calibrated population density. Because the archived raster captures lack independently verifiable cross-acquisition calibration metadata, temporal interpretation is limited to relative spatial configurations within the sampled profiles rather than absolute changes in population presence. We applied spatial mapping, Getis-Ord Gi* statistics, and Geographic Detector factor and interaction analyses. A formal block-level comparison based on within-acquisition fractional ranks showed very strong spatial similarity between the sampled workday and rest-day profiles (Pearson r = 0.980; Spearman’s rho = 0.979; both p < 0.001), indicating persistent relative spatial ordering without implying equal absolute activity levels. Archived ArcGIS sensitivity outputs further show that Gi* clustering was examined under multiple neighborhood specifications rather than a single undocumented default setting. Degree of adjacency function mixing had the highest single-factor q-values in the aggregated sampled profiles (0.461 for workdays and 0.538 for rest days), and its interaction with spatial compactness had the highest joint q-values (0.798 and 0.823). These results identify a consistent spatial association between neighboring functional context, urban form, and relative activity intensity within the observed sample. From a sustainability perspective, the findings support area-based coordination of complementary functions, accessibility, spatial form, and time-specific demand. In contrast, the temporal and spatial conclusions are deliberately bounded by the relative, block-based analytical framework and the 2020 sampling context.

Urban street-view imagery provides a scalable basis for examining visible environmental conditions relevant to human-centered sustainability. This study combines SegFormer-B5 semantic segmentation, six ResNet-50 classifiers trained using Place Pulse 2.0 annotations, spatial autocorrelation, XGBoost–SHAP, and Grad-CAM to analyze 20,256 images from 5064 sampling locations in the Macau Peninsula. We derived four objective visual indicators from the segmentation results. The mean classification accuracy across the six perception classifiers was 73.70% in the source-domain validation. The image-derived indicators and model-predicted scene categories exhibit significant spatial clustering, including global Moran’s I values of 0.673 for the Boring score and 0.637 for the Green View Index. XGBoost–SHAP further identifies nonlinear associations between 19 semantic scene components and the six model-predicted perception scores, while Grad-CAM visualizes the image regions that contribute to individual predictions. The integrated framework links streetscape measurement, spatial diagnosis, and model interpretation to support systematic identification of streets requiring detailed urban-design assessment and locally informed renewal planning.