Abstract: Persistent Scatterer Interferometric Synthetic Aperture Radar (PSInSAR) has emerged as a powerful technique for ground deformation monitoring, leveraging time-series SAR imagery to suppress phase noise caused by atmospheric delays, temporal decorrelation, and geometric distortions, extracts millimeter-scale displacement signals while preserving spatial resolution. Despite its proven effectiveness, conventional PSInSAR implementations encounter three critical bottlenecks when processing large-scale datasets. First, the spatial network partitioning strategy—typically based on regular grids or Delaunay triangulation—often fails to account for the heterogeneous distribution of persistent scatterers. Second, the parameter estimation procedure, which simultaneously solves for DEM error and linear deformation rate, relies on two-dimensional grid searches or iterative optimization. This approach becomes computationally prohibitive when processing tens of millions of PS points, as the search space expands quadratically. Third, existing quality control mechanisms are often insufficient, lacking hierarchical constraints that propagate accuracy guarantees from local arc-level measurements to global network adjustments. This can lead to error accumulation, outlier contamination. Objectives: To address these challenges, this study proposes a novel PSInSAR methodology based on image segmentation and a stepwise hierarchical strategy. The core innovations aim to transform the conventional processing chain into an adaptive, efficient, and robust framework suitable for large-scale, high-precision deformation monitoring. Methods: The proposed method consists of four key technical components, each designed to overcome a specific limitation of traditional PSInSAR. 1) Adaptive spatial network partitioning using SAR image segmentation: Instead of imposing a regular grid, we first apply an image segmentation algorithm to the amplitude of the SAR scene. The segmentation partitions the spatial domain into irregular but homogeneous regions, where each region contains PS points with similar statistical properties and spatial continuity. This segmentation-based partitioning adapts to the actual density and distribution of persistent scatterers, ensuring that each region maintains a manageable number of PS points (e.g., between 500 and 5000 points). 2) Stepwise computation strategy with dimension reduction: Traditional PSInSAR solves for both DEM error and linear deformation rate simultaneously using a two-dimensional periodogram or grid search over the parameter space. This 2D search requires evaluating numerous candidate pairs, leading to O(N × M) complexity, where N and M are the search dimensions for each parameter. Our method decouples this problem by introducing a stepwise hierarchical strategy. First, we estimate the DEM error using a subset of interferograms with short temporal baselines, where the deformation contribution is minimal. This reduces the problem to a one-dimensional search. Once the DEM error is refined, we fix it and solve for the deformation rate, again as a one-dimensional search. 3) Multi-level quality control framework: To ensure solution reliability, we stablish a multi-level quality constraint system, integrating single arc quality control and overall quality control strategy for spatial networks, to ensure the solution quality of networks at all levels. 4) Control network optimization: Finally, we implement a control network optimization procedure that enhances robustness. This includes: (a) connecting isolated PS points that were not linked during initial arc formation. (b) rapidly joining spatial subnetworks by identifying bridge points. (c) introducing redundant connections to strengthen network connectivity, making the solution less sensitive to individual arc failures. Results: We validated the proposed method using TerraSAR data from the Changsha region, the results demonstrate significant improvements over traditional PSInSAR method. Processing time was reduced to 16.05% of conventional methods. The number of monitored points increased by 29.61%. Absolute elevation error decreased from 3.128 m to 1.825 m, representing a 41.66% improvement in accuracy. The spatial continuity of processing result showed significant improvement. Conclusions: The proposed PSInSAR methodology, grounded in SAR image segmentation and a stepwise hierarchical parameter estimation strategy, effectively resolves the key bottlenecks of traditional PSInSAR—namely, unreasonable spatial partitioning, inefficient parameter solving, and inadequate quality control. By adapting the spatial network to PS heterogeneity, reducing multi-parameter searches to sequential single-parameter solutions, and enforcing multi-level accuracy constraints, the approach achieves simultaneous gains in computational efficiency, point density, measurement precision, and result spatial continuity. The experiment demonstrates strong potential for operational deployment in large-scale, high-precision deformation monitoring applications, including urban infrastructure health assessment, landslide early warning, and high-resolution DEM generation.