Ground deformation associated with rapid urbanization has emerged as a common global challenge facing cities. However, identifying its contributing mechanisms remains difficult because multiple factors, including groundwater management, engineering loads, and extreme weather events, are often superimposed in space and time. This study proposes a scale-aware analytical framework that integrates Small Baseline Subset InSAR (SBAS-InSAR), Stationary Wavelet Transform (SWT), and Independent Component Analysis (ICA) to separate deformation signals by spatial scale and statistical structure, and then support the interpretation of their potential drivers. The original deformation field is first decomposed into a large-scale background component and a small-scale local component using SWT. ICA is then applied to the large-scale background component to extract two statistically separated deformation modes with distinct spatiotemporal characteristics. These modes are further examined using Least-Squares Cross-Wavelet Analysis (LSCWA) and ancillary hydrological information to evaluate their associations with groundwater-related processes. We applied and validated this framework in Zhengzhou, China, as a representative case. Results show that Zhengzhou exhibits a clear “central uplift and peripheral subsidence” pattern. The large-scale component contains two statistically separated modes. The first mode shows a long-term uplift pattern that is spatially consistent with the groundwater no-take zone and temporally associated with confined-aquifer water-level recovery. The second mode shows an annual oscillation that is coherent with groundwater storage anomaly, with an estimated deformation lag of approximately 1–2 months. The small-scale local component highlights subsidence bowls that are spatially and temporally associated with engineering and land-use features, including newly constructed transportation corridors, high-density building areas, landfills, and waterfront park projects. The proposed framework provides a scale-aware pathway for interpreting complex urban deformation fields and may support more targeted geohazard monitoring and urban planning in areas affected by coupled hydrological and anthropogenic pressures.

Accurate petrographic identification from thin-section images is important for geoscientific analysis but remains strongly dependent on expert interpretation. In this study, we organized 2634 polarized-light images from 324 distinct thin sections representing 108 lithologies and paired them with Chinese petrographic descriptions. Four conventional vision models were trained as closed-set lithology–classification baselines, and five CN_CLIP variants were fine-tuned for image–text alignment. To prevent overlap of images from the same thin section across outer folds, thin-section-wise three-fold evaluation was performed after excluding eight lithologies represented by fewer than three distinct thin sections; this protocol therefore comprised 2530 images from 311 thin sections and 100 eligible lithologies. At the fixed 20-epoch endpoint, CN_CLIP-ViT-L/14@336px and CN_CLIP-ViT-H/14 showed bidirectional image–text matching accuracies of 81.39 ± 1.97% and 81.31 ± 0.61%, respectively, whereas the conventional classifiers achieved Top-1 accuracies of 50.37%–58.20% under their separate closed-set task. These metrics correspond to different task formulations and are not directly comparable. Under the complementary image-level 8:1:1 protocol, CN_CLIP-ViT-H/14 achieved image-to-text and text-to-image matching accuracies of 93.75% and 93.38%, respectively, on the hold-out test subset. In an inference-stage ablation using the same checkpoint, hold-out test batches, and evaluation protocol, removing only the explicit lithology name reduced these accuracies to 82.72% and 77.21%, indicating that the rock-name token is informative but not the sole source of cross-modal discrimination. Without task-specific retraining, the same CN_CLIP framework also supported constrained lithology, major-mineral, and polarization-mode matching, with best task-specific accuracies of 90.57%, 66.51%, and 86.42%, respectively. Overall, the results support vision-language learning as a flexible framework for expert-assisted petrographic image–text matching on held-out thin sections within the available dataset, while emphasizing source-aware partitioning and cautious interpretation of task-specific accuracies.